An Introduction to Fluid Dynamics

G. K. Batchelor
- Notes and Solutions

Amey Joshi

20-Jul-2014
Contents

1 The Physical Properties of Fluids 2

1.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Volume forces and surface forces action on a fluid . . . . . . . . . 2
1.3 Mechanical equilibrium of a fluid . . . . . . . . . . . . . . . . . . 3
1.3.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Classical Thermodynamics . . . . . . . . . . . . . . . . . . . . . . 13
1.5 Transport Phenomena . . . . . . . . . . . . . . . . . . . . . . . . 16
1.6 The distinctive properties of gases . . . . . . . . . . . . . . . . . 17
1.7 Conditions at a boundary between two media . . . . . . . . . . . 19
1.8 Exercises for the chapter . . . . . . . . . . . . . . . . . . . . . . . 26
1.9 Additional exercises . . . . . . . . . . . . . . . . . . . . . . . . . 28

2 Kinematics of Fluid Flow 29

2.1 Specification of the fluid flow . . . . . . . . . . . . . . . . . . . . 29
2.2 Conservation of mass . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 Analysis of the relative motion near a point . . . . . . . . . . . . 35
2.4 Velocity distribution with specified rate of expansion and vorticity 40
2.5 Singularities in the rate of expansion . . . . . . . . . . . . . . . . 42
2.6 The vorticity distribution . . . . . . . . . . . . . . . . . . . . . . 47
2.7 Irrotational and solenoidal velocity distributions . . . . . . . . . 57
2.8 Irrotational solenoidal flow in doubly-connected regions . . . . . 60
2.8.1 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.9 Three dimensional flow fields extending to infinity . . . . . . . . 62
2.10 Two dimensional flow fields extending to infinity . . . . . . . . . 70
2.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

3 Equations governing the motion of a fluid 82

3.1 Material integrals in a moving fluid . . . . . . . . . . . . . . . . . 82
3.2 Equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.3 The expression for the stress tensor . . . . . . . . . . . . . . . . . 89
3.3.1 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.4 Changes in the internal energy of a fluid in motion . . . . . . . . 93
3.5 Bernoulli’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.6 The complete set of governing equations . . . . . . . . . . . . . . 101
3.6.1 Isentropic flows . . . . . . . . . . . . . . . . . . . . . . . . 102
3.7 Incompressible flows . . . . . . . . . . . . . . . . . . . . . . . . . 104

i
4 Some mathematical results 108
4.1 Green’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.2 Solution of Poisson equation . . . . . . . . . . . . . . . . . . . . . 108
4.3 Another form of Stokes’s theorem . . . . . . . . . . . . . . . . . . 111
4.4 Multipole expansion . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.5 Solid Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.6 Green function of Laplacian . . . . . . . . . . . . . . . . . . . . . 115
4.6.1 Method 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.6.2 Method 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.6.3 Method 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.7 Divergence theorem in a plane . . . . . . . . . . . . . . . . . . . 118
4.8 Circular harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.9 A propertyR of harmonic functions . . . . . . . . . . . . . . . . . . 119
4.10 Proof of R ni nj dΩ = (4π/3)δij . . . . . . . . . . . . . . . . . . . . 120
4.11 Proof of ni nj nk nl dΩ = (4π/15)(δij δkl + δik δjl + δil δjk ) . . . . . 121
4.12 Isotropic tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Chapter 1

The Physical Properties of

Fluids

1.1 Notation
The notations in these notes differ from the book in a few cases.

• The symbol σij denotes both the tensor itself as well as its ijth component.
Sometimes, we will denote the tensor itself as {σij }.
• Vector product of a and b is written as a ∧ b. Likewise, curl of a vector
field A is denoted by ∇ ∧ A.
• We use standard notation of thermodynamics instead of the one in the
book. Thus, internal energy is U , enthalpy is H and Helmholtz free energy
is A. Further, we will follow the convention of using capital letters for
extensive quantities and small letters for intensive ones. Further, the
inexact differential of a function, say Q, is denoted by dQ
¯ while dS denotes
exact differential of S.

• We denote average of a quantity x as hxi instead of x.

• Components of velocity vector are u1 , u2 , u3 instead of u, v, w.
• The azimuthal angle in cylindrical and spherical polar coordinates is de-
noted by ϕ.

1.2 Volume forces and surface forces action on

a fluid
• The stress tensor σij is symmetric unless the fluid has a natural body
couple. We can prove it by considering the couple (torque) N on a small
volume element of the fluid, due only to surface forces. If F is the force
density then Z
N= r ∧ FdV

2
 If the volume element is small enough so that F is almost uniform over it,
the couple is of O(r4 ) = O(V 4/3 ). In terms of the stress tensor,
Z
Ni = ijk rj σkl nl dA (1.2.1)

Using divergence theorem,

Z
∂
Ni = (ijk rj σkl ) dV
∂rl
Z  
∂σkl
= ijk δjl σkl + rk dV
∂rl
Z Z
∂σkl
= ijk σjk dV + ijk rk dV (1.2.2)
∂rl
From (1.2.1) it is clear that if the volume element is small enough for σij
and ∂σkl /∂rl to vary slowly over it then Ni is O(V 4/3 ). By the same
reasoning, the first term on the right hand side of (1.2.2) is O(V ) while
the second term is O(V 4/3 ). Thus, in the limit of V going to zero, left
hand side of (1.2.2) and the second term on its right hand side vanish
faster than the first term on its right hand side. Therefore, the equation
can continue to be valid in this limit only if the first term on its right hand
side is zero, which can happen for an arbitrary volume element only if

ijk σjk = 0 (1.2.3)

– If there is a natural body couple, as in the case of a polarized fluid,

then σij will not vary slowly over the volume element. In fact, it will
end up being of the order O(V x ) for some real number x. In that
case, the right hand side of (1.2.1) is O(V 1+x ), first term on right
hand side of (1.2.2) is O(V 1+x ) and the second term is O(V 1+x ).
Since all terms are of the same order, we cannot draw any conclusion
on the nature of σij .
– We will now show that (1.2.3) implies σij = σji . Since ijk = 0 if any
two or more indices are equal, the left hand side of (1.2.3) is

(σ23 − σ32 )e1 + (σ31 − σ13 )e2 + (σ12 − σ21 )e3 = 0

The symmetry of the tensor follows immediately.

– From the preceding equation it is clear that if the stress tensor is
symmetric then the total couple on a volume element of a fluid is
zero. We will use this fact in section 1.3.

1.3 Mechanical equilibrium of a fluid

• A volume element of a fluid is in equilibrium if the total force and couple
on it vanish. We saw in section 1.2 that the symmetry of stress tensor
leads to vanishing of torque. Now, the total force on the volume element
is Z Z
ρFi dv + σij nj dA

3
For a static fluid, σij = −pδij . Using divergence theorem on the second
term,
Z Z
∂σij
σij nj dA = dV
∂xj
Z
∂pδij
= − dV
∂xj
Z
∂p
= − δij dV + 0
∂xj
Z
= − ∇ pdV

Thus, the total force on the volume element is

Z
(ρF − ∇ p) dV

It is zero for an arbitrary volume element if

ρF = ∇ p (1.3.1)

– Since ∇ p is in the direction of F, p is constant along a direction

perpendicular to it. Thus, the level surfaces of p are all normal to F.
– Equation (1.3.1) tells that a fluid is in mechanical equilibrium only for
those arrangements of ρ and F for which their product is a gradient
of a scalar field, called the pressure.
– If F is a conservative force, then (1.3.1) can be written as

− ρ ∇ Ψ = ∇ p, (1.3.2)

where Ψ is the potential of F. Taking curl of both sides,

∇ρ∧∇Ψ = 0 (1.3.3)

or that ∇ ρ is parallel to ∇ Ψ. From (1.3.2) ∇ Ψ is also parallel to ∇ p.

Thus, a fluid is in mechanical equilibrium when the level surfaces of
the three scalar fields, ρ, p and Ψ are identical.
– We will now show that (1.3.2) is equivalent to ρ = −dp/dΨ. We start
with
∂p ∂p ∂p
dp = dx + dy + dz
∂x ∂y ∂z
or,
dp ∂p ∂x ∂p ∂y ∂p ∂z
= + +
dΨ ∂x ∂Ψ ∂y ∂Ψ ∂z ∂Ψ
or,
dp ∂p 1 ∂p 1 ∂p 1
= ∂Ψ
+ ∂Ψ
+
dΨ ∂x ∂x ∂y ∂y ∂z ∂Ψ
∂z

Define
ex ey ez
Φ= ∂Ψ
+ ∂Ψ
+ ∂Ψ
∂x ∂y ∂z

4
 Therefore,
dp
= ∇p · Φ
dΨ
Further, Φ · ∇ Ψ = 1. Therefore, if we take a dot product of equation
(1.3.2) with Ψ, we get
dp
= −ρ (1.3.4)
dΨ
– In the event, ρ is independent of p, as in the case of liquids under
moderate pressures, we can integrate (1.3.4) to get p = p0 − ρΨ. If
Ψ refers to gravitational field in a laboratory, p = p0 − ρgz.
• Consider a solid body of density ρs immersed in a fluid of density ρf . It
experiences force due to earth’s gravity (f 1 ) and due to pressure of the
surrounding fluid (f 2 ). The latter force is given by
Z Z
f 2 = − p(x)ndA = − ∇ pdV,

the integral being over the surface of the body. If we imagine that the
portion of body immersed in the fluid is replaced by an equivalent volume
of fluid, keeping the fluid in mechanical equilibrium, then by equation
(1.3.1), Z
f2 = − ρf FdV,
Vi

where Vi is the volume of the immersed part of the solid. The force f 1 on
the solid is Z
f1 = ρs FdV,
Vt

where Vt is the total volume of the solid and hence the total force is
Z Z
f= ρs FdV − ρf FdV (1.3.5)
Vt Vi

This is Archimedes’s principle that the solid body immersed in a fluid

experiences a ‘buoyancy’ equivalent to the ‘weight’ of the displaced fluid.
– If the body is completely immersed in the fluid and if ρs > ρf then
there will still be a net downward force on the body. It can be
balanced either by suspending the body by a string, or if the body is
magnetic, by a magnetic field.
• Consider a fluid in a vessel rotating around the z axis. The axes are chosen
such that the gravitational force is along the negative z axis. In the frame
of reference of the rotating fluid, the potential field, Ψ, is that of gravity
and centrifugal forces. Thus,
Ω2 2
Ψ = gz − (x + y 2 )
2
The level surface of Ψ is given by Ψ = c, for some constant c. The equation
of level surface is
Ω2 2
gz − (x + y 2 ) = c
2

5
 or
Ω2 2 c
z= (x + y 2 ) +
2g g
This is the equation of an elliptic paraboloid, symmetric around the z axis
and whose bottom point is at (0, 0, c/g).

• If a solid body of density ρs can stay at rest immersed in a fluid of density

ρl then, since the net force on it is zero, from equation (1.3.5),
Z Z
ρs gdV = ρf gdV (1.3.6)
Vt Vi

If the body is at rest in a rotating fluid, then we have

Z Z
ρs g + ρs Ω2 (xex + yey ) dV − ρf g + ρf Ω2 (xex + yey ) dV = 0



Vt Vi

Using (1.3.6),
Z Z
ρs Ω2 (xex + yey )dV − ρf Ω2 (xex + yey )dV = 0
Vt Vi

Now, let us assume that the body is completely immersed in the fluid in
such a way that the net vertical force on it is zero, either by suspending
it by a string or by magnetic field. In that case, Vt = Vi and the above
equation because,
Z
(ρs − ρf )Ω2 (xex + yey )dV = 0
Vt

Since we assumed that ρs > ρf , this equation can be satisfied only if

x = y = 0, or if the body is at the center of the vessel.
• Pressure inside a self-gravitating star of uniform density. Pressure distri-
bution if given by
d r2 dp
 
= −4πGr2 ρ
dr ρ dr
If ρ = ρ0 is a constant, the equation simplifies to

d2 p 2 dp
+ = −4πGρ20 (1.3.7)
dr2 r dr
The corresponding homogeneous equation is

d2 p 2 dp
+ =0
dr2 r dr
It can be solved by first putting q = dp/dr in which case, it becomes
dq/dr + (2/r)q = 0. Its solution is q = K1 /r2 , where K1 is a constant of
integration. In terms of p(r), it can be solved further to get

K1
p0 (r) = − + K2
r

6
 where the subscript 0 indicates solution of the homogeneous equation and
K2 is another constant of integration. The particular solution p1 (r) can be
assumed to be of the form c0 +c1 r +c2 r2 where c0 , c1 and c2 are constants.
Substituting p1 (r) in equation (1.3.7), we get 6c2 + 2c1 /r + 4πGρ20 = 0.
If this equation has to be true for all r, we should have c1 = 0 and
c2 = −2/3πGρ20 . Thus the complete solution of equation (1.3.7) is

K1 2
p(r) = − + K2 − πGρ20 r2
r 3
This solution is subject to two boundary conditions:

(a) p(0) should be finite and

(b) p(a) = 0 where a is the radius of the star.
The first condition implies K1 = 0 and the second one implies K2 =
−2/3πGρ20 a2 . Thus the pressure distribution is given by
2
p(r) = πGρ20 (a2 − r2 )
3

• A polytropic process is a quasi-static process in which the specific heat of

a substance remains constant 1 Along a polytropic change of a substance
pv m = C, where p is its pressure, v is its specific volume, m and C
are constants. In terms of density ρ = v −1 , p = Cρm . If we introduce
n = (m − 1)−1 , then we have

p = Cρ1+1/n

We will solve the equation of equilibrium for a self-gravitating star

1 d r2 dp
 
= −4πGρ (1.3.8)
r2 dr ρ dr

for polytropic relation with n = 5. The solution requires a series of trans-

formations, the first of which is

ρ = λθn , (1.3.9)

where λ is, at the moment, an arbitrary constant. In terms of θ, the

polytropic relation becomes

p = Cλ(n+1)/n θn+1 (1.3.10)

and equation (1.3.8) becomes

Cλ1/n−1 (n + 1) 1 d
 
2 dθ
r = −θn (1.3.11)
4πG r2 dr dr
1 See page 40 of ‘An Introduction to the Study of Stellar Structure’, Dover Publications
(1959) by S. Chandrasekhar. This solution largely follows the one mentioned in Chan-
drasekhar’s book.

7
Introduce a new dimensionless variable ξ = r/b to transform the above
equation to,

Cλ1/n−1 (n + 1) 1 1 d
 
2 dθ
ξ = −θn
4πG b2 ξ 2 dξ dξ

If we choose
Cλ1/n−1 (n + 1)
b2 = , (1.3.12)
4πG
we get  
1 d 2 dθ
ξ = −θn (1.3.13)
ξ 2 dξ dξ
This is the Lane-Emden equation for index n. Its solutions, subject to the
boundary conditions θ(0) = 1 and θ0 (0) = 0, prime denoting differentiation
with respect to ξ, are called Lane-Emden functions of index n. A further
transformation x = 1/ξ, changes (1.3.13) to

d2 θ
x4 = −θn (1.3.14)
dx2
Equation (1.3.14) has a solution of the form θ(x) = a1 xk , where a1 is a
constant if

k+2 = nk (1.3.15)
k(1 − k) = an−1
1 (1.3.16)

which on simplification give

2
k = (1.3.17)
n−1
 1/(n−1)
2(n − 3)
a1 = (1.3.18)
(n − 1)2

Thus,
 1/(n−1)
2(n − 3) 2
θ(x) = x n−1 (1.3.19)
(n − 1)2
is a singular solution of (1.3.13). To get a general solution, we seek a
function z(x) such that θ(x) = Axk z(x) is a solution of (1.3.13). A is an
arbitrary constant. Substituting this form in (1.3.14) we get,

d2 z dz
x2 + 2kx + k(k − 1)z + An−1 z n = 0 (1.3.20)
dx2 dx
A further transformation x = et takes it to
d2 z dz
+ (2k − 1) + k(k − 1)z + An−1 z n = 0 (1.3.21)
dt2 dt
Since A was arbitrary, we choose it to be a1 . Therefore, using (1.3.16) we
get
An−1 = k(1 − k) (1.3.22)

8
Equation (1.3.21), therefore, becomes
d2 z dz
2
+ (2k − 1) − k(k − 1)z(1 − z n−1 ) = 0 (1.3.23)
dt dt
Using (1.3.17), we get
d2 z 5 − n dz 2(n − 3)
+ − z(1 − z n−1 ) = 0 (1.3.24)
dt2 n − 1 dt (n − 1)2
We now specialize to n = 5 so that
d2 z z
− (1 − z 4 ) = 0 (1.3.25)
dt2 4
or
dz d2 z z dz
= (1 − z 4 )
dt dt2 4 dt
or  2
dz z2 z6
= − + 2D,
dt 4 12
where 2D is a constant of integration. It is arbitrarily set to zero so that
dz dt
q =−
z 1− z4 2
3

We can integrate it if we put

z4
= sin2 ζ (1.3.26)
3
to get
csc ζdζ = −dt
or
− ln(csc ζ + cot ζ) = −t + α,
or  
ζ
tan = Ke−t , (1.3.27)
2
K being a constant of integration. From (1.3.26) and (1.3.27),
1/4
12K 2 e−2t

z= (1.3.28)
(1 + K 2 e−2t )2
Since we chose A = a1 ,
1/4
12K 2 e−2t

k k
θ(x) = a1 x z(x) = a1 x (1.3.29)
(1 + K 2 e−2t )2
Since we specialized to n = 5, from equations (1.3.17) and (1.3.18), we get
1
k = (1.3.30)
2
1
a1 = (1.3.31)
41/4

9
 Using (1.3.30) and (1.3.30) in (1.3.29), and noting that x = ξ −1 = et , we
get
1/4
3K 2

θ(ξ) = (1.3.32)
(1 + K 2 ξ 2 )2
Since θ(ξ = 0) = 1, 3K 2 = 1 and hence,
 1/4
1 1
θ(ξ) = = (1.3.33)
(1 + ξ 2 /3)2 (1 + ξ 2 /3)1/2

This equation is also reported on the Wolfram website. Putting n = 5 in

(1.3.10) and using the above equation in it, we get

Cλ6/5
p(ξ) = Cλ6/5 θ6 = (1.3.34)
(1 + ξ 2 /3)3

Since ξ = r/b,
27b6 Cλ6/5
p(r) = (1.3.35)
(3b2 + r2 )3
Putting n = 5 in (1.3.12),
r
2/5 3C 1
λ =
2πG b
Using this expression in (1.3.35),

27b3 C (3C)3/2
p(r) =
(3b2 + r2 )3 (2πG)3/2
√
27(b 3)3 C 5/2 1
= 3/2
√
(2πG) ((b 3) + r2 )3
2

√
Let a = b 3 so that,

27a3 C 5/2 1
p(r) = (1.3.36)
(2πG)3/2 (a2 + r2 )3

1.3.1 Exercises
1. Choose the coordinate axes such that the vessel rotates about the y axis
and the gravity is along the negative z axis. Therefore, force on unit
volume element of the fluid is

F = −gez + Ω2 (xex + zez )

The corresponding potential is

Ω2 2 Ω2 Ω2
 
2gz
Ψ(r) = gz − (x + z 2 ) = − x2 − 2
z − 2
2 2 2 Ω
or,
Ω2 2 Ω2  g 2 g2
Ψ(r) = − x − z− 2 +
2 2 Ω 2Ω2

10
 Level surfaces of Ψ(r) = −K have equation
Ω2 2 Ω 2  g 2 g2
x + z− 2 =K+
2 2 Ω 2Ω2
or
g2
 
 g 2 2
x2 + z − 2 = 2 K +
Ω Ω 2Ω2
Since the level surfaces of pressure, Ψ and ρ coincide when the fluid is in
mechanical equilibrium, above equation also describes level surfaces of p
for every value of K. Further, the equation describes a family of cylinders
whose axis is parallel to the y axis and is at a height g/Ω2 above it.
2. Density distribution is given by ρ = ρc (1 − βr2 ). Pressure distribution if
given by equation (1.3.7),
d r2 dp
 
= −4πGr2 ρ
dr ρ dr
Substituting expression for density distribution in, we get
d2 p 2 dp
2
+ 2
+ 4πGρ2c (1 − βr2 ) = 0 (1.3.37)
dr r(1 − βr ) dr
The corresponding homogeneous equation is
d2 p 2 dp
=− (1.3.38)
dr2 r(1 − βr2 ) dr
If q = dp/dr, it can be integrated as
ln |qr2 | = ln |K1 (1 − βr2 )| (1.3.39)
where K1 is a constant of integration. Second integration gives
 
1
ph (r) = −K1 + βr + K2 (1.3.40)
r
where K2 is the second constant of integration and the subscript h indi-
cates the solution of the homogeneous equation. The particular integral
pc (r) is found by the trial function
7
X
ck x7−k
k=0

Substituting this form in equation (1.3.37) and equating coefficients of like

powers of r we get
c0 = 0
2π 2 2
c1 = − Gρ β
15 c
c2 = 0
8π 2
c3 = Gρ β
15 c
c4 = 0
2π
c5 = − Gρ2c
3
c6 = 0
c7 = 0

11
Thus the solution of (1.3.37) is
2π 2 2 6 8π 2 3 2π 2 2 K1
p(r) = − Gρ β r + Gρ βr − Gρc r − − K1 βr + K2
15 c 15 c 3 r
The two boundary conditions are (1) p(0) should be finite and (2) p(a) = 0.
The first one requires K1 = 0 while the second one gives
2π 2 2 6 8π 2 4 2π 2 2
K2 = Gρ β a − Gρ βa + Gρc a
15 c 15 c 3
The complete solution of (1.3.37) thus is
2π 2 2 6 8π 2 2π 2 2
p(r) = Gρ β (a − r6 ) − Gρ β(a4 − r4 ) + Gρc (a − r2 ) (1.3.41)
15 c 15 c 3
Equation (1.3.41) is the pressure distribution in a self-gravitating spherical
star with density distribution ρ = ρc (1 − βr2 ). Mean density of the star is
Z a Z π Z 2π
3
hρi = ρc (1 − βr2 )r2 sin θdrdθdφ
4πa3 0 0 0
Solving the integral we get
 
3
hρi = ρc 1 − βa2
5
Given that the mean density of the star is twice the surface density, that
is, hρi = 2ρc (1 − βa2 ). This condition gives β = 5/(7a2 ). The total mass
of such a star is
Z a Z π Z 2π
M= ρc (1 − βr2 )r2 sin θdrdθdφ,
0 0 0

after evaluating the integral and putting β = 5/(7a2 ), we get

 3
a3

a 4π 4
M = 4πρc − = ρc a3
3 7 3 7
A star of the same mass but with uniform mass distribution will have
density ρ0 = (4/7)ρc . Pressure at the center of such a star will be
 2
2 4ρc 32
pu (r = 0) = πG a2 = πGρ2c a2
3 7 147
where we have used the second equation after (1.4.8) in the book with r =
0 and the subscript ‘u’ indicates uniform mass distribution. Substituting
β = 5/(7a2 ) in (1.3.41) and evaluating it at r = 0 we get the pressure at
the center of the star with mass distribution
52
pn (r = 0) = πGρ2c a2
147
where the subscript ‘n’ indicates non-uniform mass distribution. Clearly
pn (0) 13
=
pu (0) 8

12
1.4 Classical Thermodynamics
First law of thermodynamics states that if dQ
¯ is the amount of heat entering a
system, dW
¯ is the amount of work done on it then its internal energy changes
as
dU = dQ
¯ + d̄W
Thus, although dQ
¯ and d̄W are in general inexact differentials, their sum is not.
The first law of thermodynamics thus defines a state function, called internal
energy. If the work done (on the system) is due to compression then d̄W =
−pdV , so that dU = dQ
¯ − pdV or, equivalently
   
∂U ∂U
dQ
¯ = dU + pdV = dV + dp + pdV
∂V p ∂p V

Specific heat is c = d̄Q/dT . Since dQ

¯ is an inexact differential, the value of c
depends on the path of transformation. At constant pressure,
         
∂U ∂V ∂V ∂U ∂V
cp = +p = +p
∂V p ∂T p ∂T p ∂T p ∂T p

Similarly, specific heat at constant volume is,

     
∂U ∂p ∂U
cV = =
∂p V ∂T V ∂T V
The ratio of the two specific heats, cp /cV , is usually denoted by γ. The bulk
moduli of fluids under adiabatic (constant entropy, S) and isothermal (constant
temperature T ) are defined as
 
1 ∂p
KS = −
v ∂v S
 
1 ∂p
KT = − ,
v ∂v T
where v is the specific volume of the material. Similarly, the two compressibili-
ties are defined as
 
1 ∂v
κS = −
v ∂p S
 
1 ∂v
κT = −
v ∂p T
The proof in the textbook considers changes in Q and T as vectors in phase
diagram of the fluid. If ti and ta are unit tangents to the isothermal and
adiabatic curves and if eV and ep are unit vectors along V and p axes then,
ti = mV eV + mp ep
ta = nV eV + np ep
Let m and n be unit normals to the isotherm and adiabatic curves.
m = −mp eV + mV ep
n = −np eV + nV ep

13
If θi and θa are the angles made by these tangents with the V axis and if ti = 1
and ta = 1 are their magnitudes then

mV = cos θi
mp = sin θi
nV = cos θa
np = sin θa

Therefore,
 
mp ∂p
tan θi = = (1.4.1)
mV ∂V T
 
np ∂p
tan θa = = (1.4.2)
nV ∂V S
Further, if Thus, is δQ and δT are the ‘vector’ changes in heat and temperature
then

δQ = nV δQeV + np δQep
δT = mV δT eV + mp δT ep

If p is kept constant then

nV δQ
cp =
mV δT
Similarly, if V is kept constant then,
np δQ
cV =
mp δT
Therefore,
cp nV mp
γ= =
cV np mV
Using equations (1.4.1) and (1.4.2),
  . 
∂p ∂p
γ=
∂V S ∂V T
The same relations hold good if we use specific volume. Thus,
  . 
∂p ∂p KS κS
γ= = =
∂v S ∂v T KT κT

An easy proof of the these equations is available on Wikipedia2 .

One of the equivalent statements of second law of thermodynamics is that
the integrating factor of dQ
¯ is 1/T , where T is the absolute temperature. The
resulting exact differential is of the state function called entropy S. Thus,
dQ
¯
dS =
T
Maxwell’s relations are derived from the definition of the four thermodynamic
relations.
2 Alternatively, refer to equation 1.22 of Statistical Mechanics, 2nd edition, by K. Huang.

14
• Internal energy U is defined as dU = T dS − pdV . Thus, U can be consid-
ered as a function of S and V or that
   
∂U ∂U
dU = dS + dV
∂S V ∂V S
Therefore,

∂U
T = (1.4.3)
∂S V
 
∂U
p = − (1.4.4)
∂V S
Further, from the equality of the two second derivatives of U
   
∂T ∂p
=− (1.4.5)
∂V S ∂S V

• Enthalpy H is defined as H = U + pV . Therefore, dH = dU + pdV + V dp.

Using the relation T dS = dU + pdV , we have dH = T dS + V dp. Thus, H
can be considered as a function of S and p, or that
   
∂H ∂U
dH = dS + dp
∂S p ∂p S
Therefore,
 
∂H
T = (1.4.6)
∂S p
 
∂U
V = (1.4.7)
∂p S

Further, from the equality of the two second derivatives of H

   
∂T ∂V
= (1.4.8)
∂p S ∂S p

• Helmholtz free energy is defined as A = U − T S. Therefore, dA = dU −

T dS − SdT . Using the relation T dS = dU + pdV , dA = −pdV − SdT .
Thus, A can be considered as a function of V and T , or that
   
∂A ∂A
dA = dV + dT
∂V T ∂T V
Therefore,

∂A
p = − (1.4.9)
∂V T
 
∂U
S = − (1.4.10)
∂T V
Further, from the equality of the two second derivatives of A
   
∂p ∂S
= (1.4.11)
∂T V ∂V T

15
 • Gibbs free energy is defined as G = H − T S. Therefore, dG = dH −
T dS − SdT . Using definition of H, dG = dU + pdV + V dp − T dS − SdT =
V dp − SdT . Thus, G can be considered as a function of P and T , or that
   
∂G ∂G
dG = dp + dT
∂p T ∂T p

Therefore,
 
∂G
V = (1.4.12)
∂p
 T
∂G
S = − (1.4.13)
∂T p

Further, from the equality of the two second derivatives of G

   
∂V ∂S
=− (1.4.14)
∂T p ∂p T

If we regard S as a function of T and p,

   
∂S ∂S
dS = dT + dp
∂T p ∂p T

Now,    
∂S ∂S cp
cp = T ⇒ =
∂T p ∂T p T
therefore,  
cp ∂S
dS = dT + dp
T ∂p T
Using (1.4.14),  
cp ∂V
dS = dT − dp
T ∂T p

The coefficient of thermal expansion, β is defined as

   
1 ∂v 1 ∂V
β= =
v ∂T p V ∂T p

Therefore,
cp
dS = dT − βvdp
T
or
T dS = cp dT − βvT dp (1.4.15)

1.5 Transport Phenomena

• Equation (1.6.14) of the book is
 2  
∂S 1 kH
ρ = kH ∇T +∇· ∇T
∂t T T

16
 Integrating over a volume element,
Z Z  2 Z  
∂ 1 kH
Sρdt = kH ∇T dV + ∇· ∇T dV
∂t T T

hence,
Z Z  2 Z
∂ 1 kH
Sρdt = kH ∇T dV + ∇ T · ndA
∂t T T

Since the volume element is assumed to be thermally isolated, there is not

heat flux across the boundary. Therefore, ∇ T · n = 0 and hence,
Z Z  2
∂ 1
Sρdt = kH ∇T dV ≥ 0
∂t T

1.6 The distinctive properties of gases

• Equal volumes of all gases, under identical conditions of temperature and
pressure, contain equal number of molecules. Using the gas law pV =
nRT , one mole of all gases, at 273.15 K and 1.01235 × 105 Pa occupies
22.4 litres. Further, one mole has 6.023 × 1023 molecules. Therefore, the
number of molecules in one cubic centimeter, which is 1 milli-litre, is

6.023 × 1023 × 10−3

= 2.69 × 1019
22.4

• Volume available for each molecule in one cubic centimeter of a gas at

273.15 K and 1.01235 × 105 Pa is
1
= 3.717472 × 10−20
2.69 × 1019
cubic centimeter. If√we imagine this volume to be a little cube then the
length of its side is 3 3.717472 × 10−20 = 3.337459 × 10−7 cm. This is also
the distance between two molecules if they were to stay still along a cubic
lattice.
• The average distance a gas molecule travels before it collides with another
molecule is called the mean free path. At 273.15 K and 1.01235 × 105
Pa it is 68 nm. Refer to Wikipedia for the formula and values. 68 nm is
6.8 × 10−6 cm.
• If f (u) is the probability that a gas molecule has velocity in the neigh-
borhood du1 du2 du3 about u and N is the number density of the gas then
the number of gas molecules, per unit volume, with velocities in the same
neighborhood is N f (u)du1 du2 du3 . If we now consider an area element of
size δA and normal n then in unit time, molecules with velocities in the
neighborhood du1 du2 du3 around u, occupy a cylinder with base δA and
along n · u. Total number of such molecules is

n · uδAN f (u)du1 du2 du3

17
 If m is the mass of each molecule then the momentum flux in the same
neighborhood is
mu (n · uδAN f (u)du1 du2 du3 )
Since ρ = mN , the momentum flux across the area element is
Z ∞Z ∞Z ∞
π = ρδA un · uf (u)du1 du2 du3
−∞ −∞ −∞

The pressure is the normal component of this stress due to momentum

flux, Z ∞Z ∞Z ∞
n·π 2
p= =ρ (n · u) f (u)du1 du2 du3
δA −∞ −∞ −∞
2
But the integral is just the mean value of (n · u) . Therefore,
2
p = ρh(n · u) i

Remark. Strictly speaking, momentum flux is a tensor of rank 2. A more

rigorous derivation of the same result is,
Z ∞Z ∞Z ∞
πij = ρδA ui uj f (u)du1 du2 du3
−∞ −∞ −∞

The pressure due to it is

Z ∞ Z ∞ Z ∞
ni πij nj 2
p= =ρ (ui ni ) f (u)du1 du2 du3
δA −∞ −∞ −∞

• Equation (1.7.7) is obtained from (1.7.6) by converting p1 , p2 , p3 to “spher-

ical coordinates in momentum space”. Thus, if

dp1 dp2 dp3 = p2 sin θdpdθdφ,

the partial integration involves a gaussian.

• The probability that a gas molecule has momentum in the neighborhood

dp1 dp2 dp3 about p is
Z Z
g(p) = · · · Ce−α dq1 . . . dqs dp4 . . . dps

Since pi = mui , dp1 dp2 dp3 = m3 du1 du2 du3 , . Therefore,

Z Z
α 2
f (u) = m3 g(p) = m3 Ce− 2 mu · · · e−αF dq1 . . . dqs dp4 . . . dps ,

where F is defined in (1.7.6). Equation (1.7.8) can be derived immediately

after using (1.7.7).

18
1.7 Conditions at a boundary between two me-
dia
• Helmholtz free energy is defined as A = U − T S. Therefore, dA = dU −
T dS − SdT . In an isothermal transformation, dA = dU − T dS = −pdV .
Thus, the change in free energy is equal to the work on the system. If
A = ρ1 V1 A1 + ρ2 V2 A2 + γa, a being the area of the interface, then the
work done in an isothermal and isochoric process is dA = γda. γ is a state
function, called ‘surface tension’.
• We will now argue, using thermodynamics, that the surface tension at the
interface of a liquid and a gas is a positive quantity. It can be shown
that under isothermal and isochoric conditions, the Helmholtz free energy
is minimum 3 . If γ < 0 then dA = −|γ|da implies that the area of the
interface will increase indefinitely. Surely such a thing does not happen.
Therefore, γ is positive.
• The mass of water in a spherical drop of radius r is
4π 3
m= r ρ
3
If Lv be the latent heat of vaporization of water, the energy needed to
vaporize it is
4π 3
mLv = r ρLv
3
The surface energy of the drop is γ × 4πr2 . The two are comparable when,
4π 3 3γ
r ρLv = 4πr2 γ ⇒ r =
3 ρLv

Using the values γ = 7.35 × 10−2 N/m, ρ = 103 m3 and Lv = 2.257 × 106
J/kg, we get r = 1.000886 × 10−10 m, which is smaller than size of the
water molecule, 2.75 × 10−10 m.
• A camphor boat is propelled because dissolution of camphor reduces sur-
face tension of water. Therefore, the area where camphor dissolves is
pulled apart by those areas of the surface which are relatively uncontam-
inated.
• The equation of a curved surface can be expressed as f (x, y, z) = z −
ζ(x, y) = 0. The equation of a unit normal to it is,
∇f
n=
| ∇ f|
Now, ∇ f = −ζx ex − ζy ey + ez . Therefore, correct up to the first order in
ζx , ζy , n = −ζx ex − ζy ey + ez . Therefore, the surface force is
I
fs = γ dx ∧ n,
C
3 p.368 of Heat and Thermodynamics by Sears and Zemansky, 6th edition, McGraw Hill
Book Company, 1981

19
 where C is the boundary of a small surface in the neighborhood of O. The
integrand dx ∧ n gives the correct direction of f s . We first evaluate the
cross product,

dx ∧ n = (ex dx + ey dy + ez dz) ∧
(−ζx ex − ζy ey + ez )
= ex (dy + ζy dz) + ey (−ζx dz − dx) + ez (−ζy dx + ζx dy)

Thus,
I I 
fs = ex dy + ζy dz +
 I I 
ey −ζx dz − dx +
I I 
ez ζx dy − ζy dx

Since I I I
dx = dy = dz = 0

we have, I I 
f s = ez ζx dy − ζy dx

Using Green’s theorem,

I ZZ
(Ldx + M dy) = (Mx − Ly )dxdy

with

L = −ζy
M = ζx

we get ZZ
f s = ez (ζxx + ζyy ) dxdy (1.7.1)

• We will now show that if R1 and R2 are the principal radii of curvature
of the surface z − ζ(x, y) = 0 then
1 1
ζxx + ζyy = + (1.7.2)
R1 R2
Let (x0 , y0 , z0 ) be the center and R be the radius of curvature at a point
(x, y, z). Then the equation of the normal at the point (x, y, z) is 4

x0 − x y0 − y z0 − z R
= = =
ζx ζy −1 1 + ζx2 + ζy2
4 Seepage 224 of An elementary treatise on solid geometry by Charles Smith, MacMillan
and Company, 1907

20
 Therefore,
x0 − x = −ζx (z0 − z)
y0 − y = −ζy (z0 − z)
Differentiating these equations,
−dx = −dζx (z0 − z) + ζx dz
= −(ζxx dx + ζxy dy)(z0 − z) + ζx (ζx dx + ζy dy)
−dy = −dζy (z0 − z) + ζy dz
= −(ζxy dx + ζyy dy)(z0 − z) + ζy (ζx dx + ζy dy)
q
If κ = 1 + ζx2 + ζy2 then z − z0 = −R/κ. Therefore,

R
−dx = (ζxx dx + ζxy dy) + ζx (ζx dx + ζy dy)
κ
R
−dy = (ζxy dx + ζyy dy) + ζy (ζx dx + ζy dy)
κ
or
   
Rζxx Rζxy
1 + ζx2 + dx + ζx ζy + dy = 0
κ κ
   
Rζxy Rζyy
+ ζx ζy dx + 1 + ζy2 + dy = 0
κ κ
Therefore,
    2
Rζxx Rζyy Rζxy
1 + ζx2 + 1 + ζy2 + − + ζx ζy =0
κ κ κ
or
2

2
R + κ (1 + ζx2 )ζyy + (1 + ζy2 )ζxx − 2ζx ζy ζxy R + κ4 = 0

ζxx ζyy − ζxy
If R1 and R2 are the two solutions of this quadratic equation, then
1 1 (1 + ζx2 )ζyy + (1 + ζy2 )ζxx − 2ζx ζy ζxy
+ =
R1 R2 κ3
or, substituting for κ,
1 1 (1 + ζx2 )ζyy + (1 + ζy2 )ζxx − 2ζx ζy ζxy
+ = (1.7.3)
R1 R2 (1 + ζx2 + ζy2 )3/2
Up to first order in ζx and ζy ,
1 1
+ = ζxx + ζyy
R1 R2
We ignored higher than first power of ζx and ζy because we did the same
thing while deriving (1.7.1). It is possible to come to the same conclusion
if we do not ignore them 5 . We do that in the next point.
5 See pages 99 and 100 of A treatise on Hydromechanics, volume 1 Hydrostatics, by Besant

and Ramsey, CBS Publishers and Distributors, reprinted Indian edition, 2004

21
• Let us repeat the analysis of the point immediately before the previous
one, now without ignoring quadratic powers of ζx and ζy . We let,
−ζx ex − ζy ey + ez
n = nx ex + ny ey + nz ez = q
ζx2 + ζy2 + 1

Then,
I I 
fs = ex nz dy + ny dz +
I 
I
ey nx dz −
nz dx +
I I 
ez nx dy − ny dx

The z component of this force is

I I 
fs z = ζx dy − ζy dx ,

which using the Green’s theorem becomes,

ZZ
fs z = (nxx + nyy )dxdy

Now,
ζx ζxx (1 + ζy2 ) − ζx ζy ζxy
nx = − q ⇒ nxx = −
ζx2 + ζy2 + 1 (ζx2 + ζy2 + 1)3/2

Similarly,
ζy ζyy (1 + ζx2 ) − ζx ζy ζxy
ny = − q ⇒ nxx = −
ζx2 + ζy2 + 1 (ζx2 + ζy2 + 1)3/2

and hence,
(1 + ζx2 )ζyy + (1 + ζy2 )ζxx − 2ζx ζy ζxy
nxx + nyy =
(1 + ζx2 + ζy2 )3/2
which from (1.7.3) gives,
1 1
nxx + nyy = +
R1 R2
• From (1.7.1) it is clear that the quantity γ(ζxx + ζyy ) had dimensions of
pressure. Indeed, this is the excess pressure responsible for maintaining
the curvature of the surface. Thus, the if p1 and p2 are the pressures in
fluids 1 and 2 and if the surface of fluid 1 protrudes in fluid 2 then
p1 − p2 = γ(ζxx + ζyy )
or, from (1.7.2),  
1 1
p1 − p2 = γ + (1.7.4)
R1 R2
Note that in (1.7.4) the center of curvature lies in fluid 1.

22
 – The surface tension acts along the boundary of the fluid. But its ef-
fects are distributed along the entire surface as an ‘effective pressure’
equal to γ(R1−1 + R2−1 ), where R1 and R2 are the radii of curvature.
The radii of curvature of a surface are the radii of curvature of the
curves of intersection of the surface with any two orthogonal planes
containing the z axis.
– The jump, that is a discontinuous rise, in pressure happens when
passing towards the side of surface on which the center of curvature
lies.
– Consider a glass capillary tube dipped in a reservoir of a liquid. If the
liquid is water, then it will rise in the capillary forming a meniscus
with contact angle less than 90◦ . The center of curvature of the liquid
surface will lie in air. Therefore there will be a jump in pressure when
passing from water to air. If pl and pg are pressures in liquid and
gas,
pg − pl = γ(R1−1 + R2−1 ) > 0
Because of the drop on pressure in the liquid, water surges up the
capillary to such height as to nullify the excess. If h is the mean
height in the capillary,

ρgh = γ(R1−1 + R2−1 )

– Consider the same experiment as above but now with mercury re-
placing water. Its contact angle with glass is greater than 90◦ . The
center of curvature of the liquid surface will lie in mercury. Therefore
there will be a jump in pressure when passing from air to mercury.
If pl and pg are pressures in liquid and gas,

pg − pl = γ(R1−1 + R2−1 ) < 0

Because of the excess pressure in the liquid, mercury level dips in the
capillary to such depth (negative height) as to nullify the excess. If
−h is the mean depth in the capillary,

−ρgh = γ(R1−1 + R2−1 )

• At the interface of a gas and a liquid, if the liquid protrudes in the gas
then from (1.7.4),  
1 1
ρgz − pg = γ +
R1 R2
If the pressure of the gas is constant then,
 
1 1
ρgz − γ + = constant (1.7.5)
R1 R2
Therefore, the protrusions can be caused by surface tension if z is of the
order of r
γ
d= , (1.7.6)
ρg
which for water is 0.27 cm.

23
• To derive equation (1.9.4) of the book we note that the equation for ζ is
1 ρg 2 1
ζ + =1
2 γ (1 + ζ 0 2 )1/2

The way θ is defined in figure 1.9.2, the slope ζ 0 = tan(π/2 − θ) = cot θ.

Therefore,
1 ρg 2
ζ + sin θ = 1
2 γ
or, putting ζ = h where the slope is cot θ,
2γ
h2 = (1 − sin θ)
ρg

• We will now integrate

ρg 2 1
ζ + =1
2γ (1 + ζ 0 2 )1/2

Using (1.7.6), that is,

γ
d2 =
ρg
hence,
ζ2 1
+ =1
2d2 (1 + ζ 0 2 )1/2
Put
ζ 1
√ =
d 2 u
Hence √
0d 2 0
ζ =− 2 u
u
Or,
1 1
2
+q =1
u 2 02
1 + 2duu4
Rearranging the terms, we get
√
 
du du
d 2 √ + √ = dy
2u2 − 1 u2 2u2 − 1
or
√
Z Z 
du du D
d 2 √ − √ =y+ , (1.7.7)
2u2 − 1 u2 2u2 − 1 d
where D is a constant of integration. We will evaluate the two integrals
separately. Let,
du
I1 = √
2u2 − 1
√
Put v = u 2. Therefore,
1
Z
dv 1 1 √
I1 = √ √ = √ cosh−1 (v) = √ cosh−1 (u 2)
2 2
v −1 2 2

24
 Similarly, let Z
du
I2 = √
u2 2u2 − 1
√
If u 2 = sec v then
!
√ Z √ √
r
1
I2 = 2 cos vdv = 2 sin v = 2 1− 2
2u

Therefore equation (1.7.7) becomes

( !)
√ √ √
r
1 −1 1 D
d 2 √ cosh (u 2) − 2 1− 2 =y+
2 2u d
√
Putting back the original variable u = d 2/ζ,
  r
−1 2d ζ2 y
cosh − 4− 2 = +D
ζ d d

Using the boundary condition ζ = h at y = 0, we get

2 2
h2 ζ2
     
y 2d 2d
= cosh−1 − cosh −1
+ 4− 2 − 4− 2
d ζ h d d

• In figure 1.9.3 of the book, a liquid column protrudes in the gas. Therefore,
if pl is the pressure in liquid and pg is that in the gas, by (1.7.4),
 
1 1
∆p = pg − pl = γ +
R1 R2

We will show that the radius of curvature is a/ cos θ. Refer to 1.1, which
is an augmentation of figure 1.9.3 in the book. We first observe that

Figure 1.1: Corresponding to figure 1.9.3 in the book

∠OAB = θ, therefore, BA = OA cos θ, or if R is the radius of curvature

then a = R cos θ. Therefore, the jump in pressure is
 
cos θ cos θ 2γ cos θ
∆p = γ + =
a a a

25
 where ∆p = pg − pl . Since pg is the atmospheric pressure, this indicates
a pressure deficit in liquid. It is compensated by supporting a column of
height H. Therefore,
∆p = ρgH
so that
2γ cos θ
ρgH =
a
or
γ cos θ
H=2
ρg a
Using (1.7.6),
2d2 cos θ
H=
a
• In the derivation of equation (1.9.7) in the book, consider a ‘gaussian
pillbox’ of the kind used in electrodynamics.

• In the analysis related to figure 1.9.4 the center of curvature is in the upper
fluid. Therefore, the pressure rises while passing from the lower fluid to
the upper. Or,  
0 00 1 1
σij nj − σij nj = γ + ni (1.7.8)
R1 R2

1.8 Exercises for the chapter

1. We will first show that the excess pressure inside a bubble or radius r is
4γ/r, where γ is the surface tension between liquid of which the bubble is
formed. When a bubble of a liquid is created, its surface tension will try
to minimize its surface area thereby compressing the gas inside. It will
keep compressing the gas until a static equilibrium between the pressure
of the gas and surface tension is reached. Let us suppose that a bubble
has reached this equilibrium stage. Imagine the bubble to be cut into
two hemispheres. The upper hemisphere exerts a surface tension force of
2 × γ × 2πr, where 2πr is the perimeter of the hemisphere and we are
multiplying by a factor of 2 because there are two surfaces in a bubble,
the outer and the inner. The upper hemisphere also exerts a pressure on
the lower hemisphere. If p is the excess pressure inside the bubble and
p0 is the ambient pressure, the force due to pressures is (p − p0 ) × πr2 .
Under the condition of equilibrium, the pressure and the surface tension
forces balance each other and we get an expression for the excess pressure
(p − p0 ) inside a bubble as 4γ/r.
Remark. If the bubble is replaced by a drop, the corresponding excess
pressure will be half that of the bubble because a drop has only one surface
unlike two in a bubble.
First law of thermodynamics is dU = dQ ¯ + d̄W . In the process of coa-
lescence of bubbles, there is no heat exchange, that is dQ¯ = 0. Further,
dW
¯ = dpV ¯ = V dp + pdV . Thus the energy change in coalescence of bub-
bles is dU = V dp + pdV . If p0 is the ambient pressure, the excess pressure

26
 inside a bubble of radius a1 is
4γ
p = p0 +
a1
and that inside a bubble of radius a2 is
4γ
p = p0 +
a2
Thus,
V dp = (4γ/a1 ) × (4πa31 /3)
for a bubble of radius a1 . As the bubble grows from a volume 0 to 4πa31 /3
under ambient pressure, the term pδV for it is p0 4πa31 /3. The energy
needed to create the bubble of radius a1 is
16π 2 4π 3
dU1 = γa1 + a p0 (1.8.1)
3 3 1
Similarly, energy needed to create a bubble of radius a2 is
16π 2 4π 3
dU2 = γa2 + a p0 (1.8.2)
3 3 2
and the energy needed to create the coalesced bubble of radius r is
16π 2 4π 3
dU = γr + r p0 (1.8.3)
3 3
Since at the end of the process of coalescence the temperature returns to
that before, dU = dU1 + dU2 . From (1.8.1), (1.8.2) and (1.8.3),

p0 r3 + 4γr2 = p0 (a31 + a32 ) + 4γ(a21 + a22 )

2. If there is no wetting, none of the liquid will stick to the sphere. The
sticking happens because of a drop in pressure in the liquid. Since there is
no more than a drop of liquid available, it forms a plano-concave lens with
the plane along the rigid surface and concave portion along the spherical
surface. The radius of curvature of the curved part of the plano- concave
lens is a. That of the plane part is infinity. Therefore, the interfacial
pressure is  
1 γ
pl − pg = γ 0 + = ,
a a
This excess pressure is distributed over the entire surface of the sphere,
resulting in an adhesive force of
γ
4πa2 × = 4πγa
a

3. The problem asks us to explain apparent attraction and repulsion of float-

ing bodies as a result of their wetting properties. Recent research shows
that wetting properties 6 are not sufficient to explain the phenomenon.
6 The ”Cheerios Effect”, D. Vella and L. Mahadevan, Am. J. Phys. 73, September 2005

27
 An explanation that just takes into account the contact angle runs as
follows. Imagine two floating bodies that have a wetting interface. Wetting
causes the surface of water to rise near the interface. Since the bodies
prefer to float and are encouraged to do so by buoyancy, a neighboring
body climbs up the rising interface giving an appearance of attraction.
Now let the two bodies have a non-wetting interface. The surface of water
is now depressed and a neighboring floating body that also has a non-
wetting interface, falls along the depressed water surface once again giving
appearance of attraction.
If one body, say B1, has a wetting interface and another, say B2 has
a non-wetting interface, then the water surface is depressed around B2
and elevated around B1. If B1 approaches B2, it prefers to slide up the
interface due to its own buoyancy giving an appearance of repulsion.

1.9 Additional exercises

1. Adhesive force between two plates wetted with a liquid. Consider two plates
with a small amount of liquid between them. If the liquid wets the plates,
it will form a convex meniscus of the type shown in figure 1.2 Since the

Figure 1.2: Plates with a wetting liquid between them

center of curvature lies in the air, the excess pressure is in the air. Equiv-
alently, there is a drop of pressure in the liquid. There is no more liquid
available to rush in the cavity to compensate the drop. Further, since
the pressure is isotropic, there is a drop in pressure between the plates,
resulting in a strong adhesion. At surface S1 , the drop in pressure is

δp = γ(R1−1 + R2−1 ),

where R1 = R2 = a/ cos θ, a being the separation between the plates. If

A is the area of the plate, then the force of adhesion is
γA cos θ
F =
a
Note that the same pressure drop is sufficient to sustain the surface S2 .
We do not have to add it to the above expression.

28
Chapter 2

Kinematics of Fluid Flow

2.1 Specification of the fluid flow

• Eulerian description of a fluid flow is similar to the field description in
electrodynamics. We specify various dynamical variables as functions of
x and t. The flow is described by the Eulerian velocity u(x, t).
• Lagrangian description of a fluid flow is similar to the particle description
in mechanics. We identify a small enough element of fluid by its center of
mass a and describe the fluid flow using the Lagrangian velocity v(a, t).
In what follows, the Eulerian description is used almost exclusively.
• A stream line is an imaginary line in the fluid whose tangent at any point is
parallel to the (Eulerian) velocity at that point. Let the dx be an element
of the stream line. That is, dx = dx1 e1 + dx2 e2 + dx3 e3 . If u is the
velocity at this element, u ∧ dx = 0, because the two vectors are parallel.
Therefore, if u = u1 e1 + u2 e2 + u3 e3 ,

e1 (u2 dx3 − u3 dx2 ) + e2 (u3 dx1 − u1 dx3 ) + e3 (u1 dx2 − u2 dx1 ) = 0

Therefore, each component is zero, or

dx1 dx2 dx3
= =
u1 u2 u3
If the flow is steady, that is u is independent of t then the stream lines at
any point are unchanged over any time interval. Further, if we consider
the set of all stream lines passing through all points of a closed curve then
they form what is called a stream tube. A stream tube carries the same
mass of fluid at any point. For supposing this was not true then at some
point along the stream tube either mass gets in it or goes out of it. At
such a point, the velocity vector is not tangent to the stream line, which
cannot happen.
• A path line is the trajectory taken by a selected fluid element. A streak line
associated with a fixed point x0 in the fluid is path taken by all elements
of the fluid passing through x0 . For example, imagine a gas let out of a
container opening along the positive x1 axis. The path lines of the gas

29
 and the streak lines will be along the positive x1 axis. If the container is
now rotated to open along positive x2 axis, the streak lines of the new gas
elements getting out the container will change. However, the path lines of
elements already out will remain unchanged. The following figure, taken
from Professor Paulo Lozano’s lectures, explain the concepts quite clearly.
In a steady flow, the stream lines, path lines and streak lines coincide.

Figure 2.1: Path lines and streak lines

• A flow field u(x, t) is said to be two dimensional if it is unchanged by

moving along one of the three axes. For example, flow along x1 axis be-
tween two infinite plates along the planes x3 = 0 and x3 = 1 is unchanged
if we move along the x2 axis. Such a flow field is two dimensional. A
flow of a fluid in a cylindrical pipe is not two dimensional. A flow field
u(x, t) is said to be axi-symmetric if, expressed in cylindrical coordinates,
is independent of ϕ.
• Consider a steady flow in a Venturi tube. Every fluid element, as it ap-
proaches the constricted portion, accelerates and then decelerates after it
passes through it. The change in its velocity is solely due to the differ-
ent velocity field prevalent near the constriction. The acceleration and
the subsequent deceleration both happen in a steady flow. This change is
captured by the concept of a material derivative. Thus,
∂u
u(x + δx, t + δt) = u(x, t) + δx · ∇ u + δt + O(δt2 ) + O(δx2 )
∂t
∂u
= u(x, t) + δtu · ∇ u + δt + O(δt2 )
 ∂t 
∂u
= u(x, t) + δt + u · ∇ u + O(δt2 )
∂t
In general, if θ is any parameter associated with the flow,
 
∂θ
θ(x + δx, t + δt) = θ(x, t) + δt + u · ∇ θ + O(δt2 )
∂t
and we define the material derivative of θ as
Dθ ∂θ
= + u · ∇θ
Dt ∂t
The first term on the right hand side is called the local rate of change of θ
at a point x. It is zero under steady conditions. The second term on the
right hand side is called the convective rate of change.

30
2.2 Conservation of mass
• Volume of material element can be written as
Z
τ = dl · ndS

If each area element ndS moves by a distance δl then the volume becomes,
Z
τ + δτ = d(l + δl) · ndS

If u is the velocity of the surface element, δl = uδt so that

Z Z Z
τ + δτ = d(l + uδt) · ndS = dl · ndS + δt du · ndS

or, Z
δτ = δt du · ndS
or Z Z
dτ
= du · ndS = ∇· udV
dt
In the limit τ → 0, we can approximate the right hand side as τ ∇· u and
hence,
1 dτ
lim = ∇· u (2.2.1)
τ →0 τ dt

• A strict definition of an incompressible fluid is the one whose density does

not get affected by changes in pressure. Density of a fluid may change be-
cause of heat conduction (thermal expansion) or changes in concentration
of solute. However, such changes are not indicative of compressibility of
the fluid. In absence of temperature and/or concentration gradients, an
incompressible fluid is the one whose density remains unchanged. Thus,
if ρ is the density of a fluid element, an incompressible fluid has
Dρ
=0
Dt
or, equivalently ∇· u = 0. The equivalent follows from equation of conti-
nuity (mass conservation).
• A stream tube ends when the velocity of the fluid drops to zero. In an
incompressible fluid, this can happen only at the boundaries with solids.
Therefore, a stream tube is either closed, or ends on a solid boundary or
extends to infinity.
• Let us apply the equation of conservation of mass,
∂ρ
+ ∇· (ρu) = 0, (2.2.2)
∂t
to two circumstances
– If the fluid is incompressible, then ρ is a constant and hence ∇· u = 0,

31
 – If the fluid is compressible, but the flow is steady. Therefore ρt = 0
and hence ∇· (ρu) = 0.
If we now assume that the flow is two dimensional, then ∇· u = 0 implies
∂u1 ∂u2
+ = 0,
∂x1 ∂x2
or u1 dx2 + (−u2 )dx1 is an exact differential and hence can be written as
dψ, in which case,
∂ψ
u1 =
∂x2
∂ψ
u2 = −
∂x1
Since dψ = u1 dx1 − u2 dx2 , we have,
Z Z
ψ − ψ0 = dψ = (u1 dx1 − u2 dx2 )

This equation is interpreted as a flux if we consider an open surface formed

by translating the curve OP , parallel to the z axis, by a unit distance.
That is, the flux is
Z 1 Z 1Z
(ψ − ψ0 )dx3 = (u1 dx1 − u2 dx2 )dx3 ,
0 0

which after integration gives the previous equation.

• We will now argue why ψ is a constant along a stream line. We first
observe that since a stream line is everywhere tangential to u, there is
no flux across a stream line. But since flux is ψ − ψ0 . Along a stream
line, ψ = ψ0 , or that ψ is unchanged. Therefore, ψ is called the stream
function.
• For a flow described in polar coordinates, ∇· u = 0 implies,
1 ∂(rur ) 1 ∂uθ
+ =0
r ∂r r ∂θ
or rur dθ − uθ dr is an exact differential, say dψ, and hence,
∂ψ
rur =
∂θ
∂ψ
uθ = −
∂r

• In the case of an axisymmetric flow, ∇· u = 0 implies,

∂ux 1 ∂(σuσ )
+ =0
∂x σ ∂σ
This can as well be written as,
1 ∂(σux ) 1 ∂(σuσ )
+ =0
σ ∂x σ ∂σ

32
 or that σux dσ − σuσ dx = dψ, an exact differential. Therefore,
∂ψ
σux =
∂σ
∂ψ
σuσ = −
∂x
or
1 ∂ψ
ux =
σ ∂σ
1 ∂ψ
uσ = −
σ ∂x
Since dψ = σux dσ − σuσ dx, we have,
Z Z
ψ − ψ0 = dψ = (σux dσ − σuσ dx)

This equation can be interpreted as a flux if we consider an open surface

formed by rotating the curve about the x axis by 2π (or any other angle
in [0, 2π]). Thus, the flux is,
Z 2π Z 2π Z
(ψ − ψ0 )dθ = (σux dσ − σuσ dx)dθ,
0 0

which after integration gives the previous equation.

2.2.1 Exercises
1. We use the fact that the volume of a tetrahedron with points A, B, C
and D is
1
V = (A − D) · (B − D) ∧(C − D)
6
Consider a small tetrahedron P0 Q0 R0 T0 around a material point P0 . Let
the coordinates of the four points be X(0) , X(0) + δX(1) , X(0) + δX(2) and
X(0) + δX(3) . Then, its volume is
1 (1)  (2)  
rst (3)
δV = δX · δX ∧ δX(3) = δXr(1) δXs(2) δXt
6 6
Let a deformation take this material tetrahedron to P1 Q1 R1 T1 around
the material point P1 . Let the coordinates of the four points be x(0) ,
x(0) + δx(1) , x(0) + δx(2) and x(0) + δx(3) . Then, its volume is
1 (1)  (2)  
ijk (1) (2) (3)
δv = δx · δx ∧ δx(3) = δxi δxj δxk
6 6

If X(0) = (X1 , X2 , X3 ) and x(0) = (x1 , x2 , x3 ) then xi = xi (X1 , X2 , X3 ).

Therefore,
∂xi ∂xi ∂xi ∂xi
dxi = dX1 + dX2 + dX3 = dXr
∂X1 ∂X2 ∂X3 ∂Xr

33
 Therefore,

(1) ∂xi
δxi = δXr(1)
∂Xr
(2) ∂xj
δxj = δX (2)
∂Xs s
(3) ∂xk (3)
δxk = δX
∂Xt t

Note that, in the above relations, the coordinates of δx(1) depends only
on those of δX(1) and so on. Thus,
ijk ∂xi ∂xj ∂xk (3)
δv = δXr(1) δX (2) δX
6 ∂Xr ∂Xs s ∂Xt t
or
ijk ∂xi ∂xj ∂xk (3)
δv = δX (1) δXs(2) δXt
6 ∂Xr ∂Xs ∂Xt r
We now use the result that for a tensor {Tij },
ijk
mpq det({Tij }) = Tim Tjp Tkq
6
so that
ijk ∂xi ∂xj ∂xk ∂(x1 , x2 , x3 )
= rst
6 ∂Xr ∂Xs ∂Xt ∂(X1 , X2 , X3 )
and hence,
∂(x1 , x2 , x3 ) (3)
δv = rst δX (1) δXs(2) δXt
∂(X1 , X2 , X3 ) r
or,
∂(x1 , x2 , x3 ) (3)
δv = rst δXr(1) δXs(2) δXt
∂(X1 , X2 , X3 )
or,
∂(x1 , x2 , x3 )
δv = δV
∂(X1 , X2 , X3 )
If the mass in the two material elements is the same, then

m ∂(x1 , x2 , x3 ) m
=
δv ∂(X1 , X2 , X3 ) δV

If ρ0 = m/V and ρ = m/v 1

∂(x1 , x2 , x3 ) ρ0
=
∂(X1 , X2 , X3 ) ρ
1 This solution is adapted from the development of mass conservation in A. J. M. Spencer’s

Continuum Mechanics, Dover publications, 1980.

34
2.3 Analysis of the relative motion near a point
• The continuum hypothesis allows us to assume that the velocity u is a
continuous function of x. Therefore, if u(x, t) is the velocity of the fluid
at a point x then the velocity, at the same time instant, at a neighboring
point x + r is,

u(x + r, t) = u(x, t) + r · ∇u + O(r2 )

The difference in velocity δu can be written in Cartesian tensor form,

∂ui
δui = rj
∂xj

It can be further manipulated as,

   
rj ∂ui ∂uj rj ∂ui ∂uj
δui = + + − ,
2 ∂xj ∂xi 2 ∂xj ∂xi

We define the rate of strain tensor eij and the vorticity tensor ξij as,
 
1 ∂ui ∂uj
eij = +
2 ∂xj ∂xi
 
1 ∂ui ∂uj
ξij = −
2 ∂xj ∂xi

• The symmetric tensor eij can be diagonalized by choosing an set of co-

ordinate axes, appropriately rotated with respect to the original ones. It
then takes a form  
a 0 0
0 b 0
0 0 c
The term ri eij , in this set of axes becomes,
  0
a1 0 0 r1
 0 a2 0  r20  = a1 r10 + a2 r20 + a3 r30
0 0 a3 r30

The contribution a1 r10 +a2 r20 +a3 r30 to δui means that the fluid elements at
x and x+r, which could have been identical if the fluid werenot in motion,
now differ in shape. The one at x+r is elongated by a factor of ai along the
ri axis. Thus a fluid element that is initially spherical becomes ellipsoidal.
If the fluid is incompressible, the volume of the ellipsoid is same as that
of the sphere. If the fluid is compressible, then we can decompose eij as
   
1 0 0 e11 − (eii /3) e12 e13
eii 
0 1 0 +  e21 e22 − (eii /3) e23 ,
3
0 0 1 e31 e32 e33 − (eii /3)

where the first (isotropic) term contributes to change in volume while the
second (deviatoric) term contributes to pure straining motion.

35
• The anti-symmetric tensor ξij is
 
0 ∂2 u1 − ∂1 u2 ∂1 u3 − ∂3 u1
∂1 u2 − ∂2 u1 0 ∂3 u2 − ∂2 u3  ,
∂3 u1 − ∂1 u3 ∂2 u3 − ∂3 u2 0

where the notation ∂i uj means

∂uj
∂xi
We observe that it has only three independent elements, (unlike 6 in the
case of eij ) and hence can be written as ξij = −ijk ωk /2, where ωk are
components of the local vorticity vector ω. Therefore, the contribution to
δu is
1
rj ξij = − ijk rj ωk ,
2
which in Gibbs form is
1 ω
− r∧ω = ∧r
2 2
This is same as the velocity of a rigid body rotating with an angular
velocity ω/2. (Recall the relation v = ω ∧ r between tangential velocity,
v and rotational velocity ω, in a uniform circular motion.)
• Since ω = ∇ ∧ u, the vorticity is twice the effective angular velocity of the
fluid. We explain it as follows. Consider a small, circular loop of radius
a around x. The tangential component of velocity averaged along the
circumference of the loop is
I
1
u · dr
2πa
Dividing further by a,
∇∧u
I Z
1 1 1
u · dr = ∇ ∧ u · ndA ≈ ∇ ∧ uπa2 =
2πa2 2πa2 2πa2 2

• Consider a spherical element of fluid at x + r. It has a velocity

u(x + r) = u(x) + r · ∇u + O(r2 )

Therefore, it has an angular momentum,

Z
L = r ∧ (u + r · ∇u) ρdV, (2.3.1)

about the point x. In Cartesian tensor notation, it is

Z  
∂uk
Li = ijk rj uk + rl ρdV (2.3.2)
∂xl
Consider the first term on the right hand side,
Z
L1 = r ∧ uρdV

36
 If the fluid element is small, u and ρ are almost constant throughout it.
Therefore, we have, Z
L1 = −ρu ∧ rdV

Since r = xex + yey + zez , when expressed in spherical polar coordinates,

rdV is
(r sin θ cos ϕex + r sin θ sin ϕey + r cos θez ) r2 sin θdrdθdϕ
or,
Z R Z π Z 2π
3 2
L1 = −ρu ∧ ex r sin θdrdθ cos ϕdϕ +
0 0 0
Z R Z π Z 2π
−ρu ∧ ey r3 sin2 θdrdθ sin ϕdϕ +
0 0 0
Z π Z R
−2πρu ∧ ez r3 dr sin θ cos θdθ
0 0
Z π
πρR4 sin 2θ
= 0+0− u ∧ ez dθ
2 0 2
= 0
The second term on the right hand side of (2.3.1) can be best manipulated
in Cartesian tensor form,
Z
∂uk
L2i = ijk rj rl ρdV
∂xl
Once again, under the assumption of a small fluid element,
Z
∂uk ∂uk
L2i = ijk rj rl ρdV = ijk Ijl , (2.3.3)
∂xl ∂xl
where Iij is the moment of inertia tensor of the fluid element about any
axis passing through the center.
• It can be shown that for a spherical fluid element,
Z
Iij = ρrj rl dV,

is a diagonal tensor. This is because the ‘product of inertia’ terms Ixy , Iyz
and Izx are all zero. We will show the proof for one of them, the rest are
similar.
Z
Ixy = xyρdV
Z R Z π Z 2π
= r4 sin3 θ sin φ cos ϕdrdθdϕ
0 0 0
Z R Z π Z 2π
4 3 sin 2ϕ
= r sin θdrdθ dϕ
0 0 0 2
Z R Z π
= r4 sin3 θdrdθ × 0
0 0
= 0

37
 Thus,
Ixy = Iyz = Izx = 0 (2.3.4)

• We will next show that the ‘moment of inertia’ terms Ixx , Iyy , Izz are all
equal to I/2, where I is the moment of inertia of the sphere through any
axis passing through the center. Thus,
Z
Ixx = x2 ρdV
Z R Z π Z 2π
= ρ r4 sin3 θ cos2 ϕdrdθdϕ
0 0 0
Z R Z π Z 2π
4 3
= ρ r dr sin θdθ cos2 ϕdϕ
0 0 0
5
R 4
= ρ π
5 3
4π
= ρ R5
15
M R2
= ,
5
where M is the mass of the fluid element. We can similarly show that

M R2
Iyy = Izz =
5
Now, the moment of inertia of a sphere of mass M and radius R about
any axis passing through the center is I = 2M R2 /5. Therefore,

I
Ixx = Iyy = Izz = (2.3.5)
2

• From equations (2.3.4) and (2.3.5),

I
Ijl = δjl
2
Therefore, from (2.3.3),

1 ∂uk 1
L2i = ijk I = ωi I,
2 ∂xj 2

which, in Gibbs notation, is

1
L2 = Iω
2
Since L1 = 0, we observe that the total angular momentum of the spherical
fluid element is,
1
L = Iω
2
This is the angular momentum the spherical element would have had if it
was rotating as a rigid body with angular velocity ω/2.

38
• Since
u(x + r) = u(x) + r · ∇ u
or
ui (x + r) = ui (x) + rj eij + rj ξij
We can write rj eij = ∂Φ/∂xi , where
1
Φ= ri eij rj
2
We also saw that
1
rj ξij = ijk ωj rk
2
Therefore,
ω
u(x + r) = u(x) + ∇ Φ + ∧r (2.3.6)
2
The three terms on the right hand side are contributions due to
– Uniform translation
– Pure straining motion
– Rigid body rotation
• We will show that for a simple shearing motion, the principal rates of
strain are
1 ∂u1 1 ∂u1
,− ,0
2 ∂x2 2 ∂x2
Since
∂u1
δu = r2 e1 ,
∂x2
the rate of strain tensor is,
   
0 a 0 0 1 0
a 0 0 = a 1 0 0
0 0 0 0 0 0
where
1 ∂u1
a=
2 ∂x2
Using the secular equation for the tensor eij , we readily observe that its
eigenvalues are a, −a, 0. Therefore, the diagonalized form of the tensor is
 
a 0 0
0 −a 0
0 0 0
Thus, the principal rates of strain are a, −a, 0 or,
1 ∂u1 1 ∂u1
,− ,0
2 ∂x2 2 ∂x2
Further, the eigen vectors corresponding to these eigenvalues are
     
1 −1 0
a a
√ −1 , √  1 , 0
2 0 2 0 1

39
 Eigenvalues and eigenvectors can be calculated in R using the following
code.
Listing 2.1: Computing eigenvalues and eigenvectors
e <- matrix ( c (0 , 1 , 0 , 1 , 0 , 0 , 0 , 0 , 0) , 3 , 3 , byrow = T )
eigen ( e ) $ values
eigen ( e ) $ vectors

• From equation (2.3.6) it is we observe that any relative velocity field can
be written as
ω
δu = u(x + r) − u(x) = ∇ Φ + ∧ r,
2
which is a sum of pure straining motion and a rigid body rotation. The
pure straining motion can itself be written as a symmetrical expansion (or
contraction) and a simple shearing motion. This follows from the fact that
any tensor can be written as a sum of the isotropic tensor and a deviatoric
tensor. For example,
e11 − e3ii
     
e11 e12 e13 1 0 0 e12 e13
e
e21 e22 e23  = ii 0 1 0 +  e21 e22 − e3ii e23 
3
e31 e32 e33 0 0 1 e31 e32 e33 − e3ii
The first term on the right hand side describes a symmetrical expansion
(or contraction) while the second term is describes pure shear. Pure shear
keeps the volume unchanged. This is because, the dilatation is given by
trace of the strain tensor 2 . Therefore, the trace of rate of strain tensor
will give the rate of dilatation. Since the trace of the second tensor on
the right hand side is zero, we conclude that it keeps the volume of a fluid
element unchanged.

2.4 Velocity distribution with specified rate of

expansion and vorticity
• We saw in (2.2.1) that the divergence of the velocity field is the local rate
of expansion of the fluid. Further, the curl of the velocity field was defined
to be the local vorticity. The goal of this section is to determine u, given
∆ = ∇· u and ω = ∇ ∧ ω.
• Let us begin with finding the velocity ue when divergence is given and curl
is zero. The subscript ‘e’ denotes that the rate of expansion is specified.
Since ∇ ∧ ue = 0, there is a scalar function φ(x) such that ue = ∇ φ.
Therefore, ∇· ue = ∆(x) ⇒ ∇2 φ = ∆(x). We show in point 4.2 of
appendix that the solution of Poisson’s equation ∇2 φ = ∆(x) is
Z ∞
1 ∆(x0 )
φ(x) = − dV 0
4π −∞ |x − x0 |
Therefore,
∞
x − x0
Z
1
ue = ∇ φ = ∆(x0 )dV 0 ,
4π −∞ |x − x0 |3
2 Refer to the Wikipedia article on Volumetric strain

40
 where we have used the result,

1 x − x0
∇ 0
=−
|x − x | |x − x0 |3

This expression is similar to that of electrostatic field (in gaussian units)

due to a volume charge density.

• We interpret
x − x0 ∆(x0 )
δue (x) = δV 0
|x − x | 4π|x − x0 |
0 2

as the irrotational velocity distribution “due to” a source ∆(x0 ) in a volume

element δV 0 around the point x0 . If we take the divergence of this equation,

∆(x0 )δV 0 x − x0
∇· δue (x) = ∇· = ∆(x0 )δV 0 δ(x − x0 )
4π |x − x0 |3

where we have used the result3

x − x0
∇· = 4πδ(x − x0 )
|x − x0 |3

The velocity field δue has a non-zero rate of expansion only at x0 . We can
interpret it like the velocity field due to a ‘point source’.

• Let us begin with finding the velocity uv when curl is given and diver-
gence is zero. The subscript ‘v’ denotes that the vorticity is specified.
Since ∇· uv = 0, there is a vector function B(x) such that uv = ∇ ∧ B.
Therefore,
ω = ∇ ∧ uv = ∇ ∧ ∇ ∧ B = ∇ (∇· B) − ∇2 B
We can choose the vector B such that its divergence is zero4 , so that

ω = −∇2 B

which is same as
∇2 Bi = −ωi ,
We once again have Poisson equation, whose solution we found to be
Z ∞
1 −ωi (x0 ) 0
Bi (x) = − dV
4π −∞ |x − x0 |
or ∞
ω(x0 )
Z
1
B(x) = dV 0
4π −∞ |x − x0 |
This equation is analogous to the expression for magnetic vector potential
due to a given current density. Further,
Z ∞
1 1
uv = ∇ ∧ B = ω(x0 ) ∧ ∇ dV 0
4π −∞ |x − x0 |
3 See chapter 1 of ’Introduction to Electrodyamics’ by David Griffiths.
4 This is analogous to Coulomb gauge in electrodynamics.

41
 or
∞
x − x0
Z
1
uv = ω(x0 ) ∧ dV 0
4π −∞ |x − x0 |3
Z ∞
1 (x − x0 ) ∧ ω(x0 ) 0
= − dV
4π −∞ |x − x0 |3
Despite its analogy with the expression for the magnetic field produced due
to a given current density, there is no ‘cause and effect’ relation between
uv and ω. In the context of fluid dynamics, this is just a kinematical
relation.
• Defining s = x − x0 , we can interpret the above result as the sum of
contribution from different volume elements of the fluid, each element
contributing
s ∧ ω(x0 ) 0
δuv = − δV
4πs3
However, unlike the case of δue , this cannot be velocity ‘due to’ a vortex
of strength ω in volume δV 0 around x0 because such a vorticity field will
not have a zero divergence.
• If u is a velocity field with a given divergence and curl then it can be
written as u = ue + uv + v, where v is solenoidal as well as irrotational.
It is determined by the boundary conditions imposed on u.

2.5 Singularities in the rate of expansion

• The velocity ‘due to’ a given rate of expansion ∆ is
Z ∞
1 x − x0
ue = ∇ φ = ∆(x0 )dV 0
4π −∞ |x − x0 |3
Let the function ∆ be such that it is non zero only in a small volume 
around the point x0 . In that case, the above integral can be approximated
as, Z
1 s
ue = ∆(x00 )dV (x00 ),
4π s3 
where s = x − x0 . If we now allow  to shrink as small as possible while
leaving the value of the integral
Z
∆(x00 )dV (x00 )


constant, say m, then we have,

1 ms
ue =
4π s3
Thus the velocity can be attributed to a ‘point source’ of strength m. For
such a point source, the velocity potential is
1 m
φ=−
4π s
The two preceeding equations are analogous to those for electric field and
electric potential due to a point charge, if we use gaussian units.

42
• The remark that the flow on one side of a rigid plane containing a small
hole through which fluid is sucked at the rate M units of volume per
second is approximately the same as the flow produced in an unbounded
fluid by a point source of strength −M is not clear to me.
• A source doublet at a point x0 is a point source of strength m at x0 +δx0 /2
and a sink of strength −m at x0 − δx0 /2 such that
µ = lim
0
mδx0 ,
δx →0

remains finite. This is analogous to the definition of a point electric dipole.

The potential at x due to a source doublet at x0 is
 
m 1 1
φe (x) = lim −
δx0 →0 4π |x − x0 + δx0 /2| |x − x0 − δx0 /2|
If we write s = x − x0 then
1 1 1
= =√ ,
|x − x0 0
+ δx /2| 0
|s + δx /2| s + s · δx0
2

where the last expression is true up to first order in δx. Thus,

−1/2
s · δx0

1 1
= 1+
|x − x0 + δx0 /2| s s2
Once again, using Taylor expansion to first order in δx,
1 s · δx0 1 1 s · δx0
 
1 1
0 0
= 1 − 2
= −
|x − x + δx /2| s 2 s s 2 s3
Similarly,
1 1 1 s · δx0
= +
|x − x0 − δx0 /2| s 2 s3
and hence,
m s · δx0
φe (x) = − lim
0δx →0 4π s3
Since  
1 s
∇ =− 3
s s
we have,  
µ 1
φe (x) = ·∇
4π s
• To get the velocity due to a source doublet, we use the results,
∇s = e1 e1 + e2 e2 + e3 e3
 
1 s
∇ = −3 5
s3 s
so that
1  µ µ·s 
ue (x) = − 3 +3 5 s
4π s s
Once again, this expression is similar to electric field at x due to a point
dipole at x0 .

43
• If µ is along the z-axis then ue has no dependence on φ (in both cylindrical
and spherical polar coordinates). That is what is meant by the statement
that ue is axi-symmetric.

• If x0 = 0, s = x. Further, the radial component of the velocity is ue · s =

ue · x. Therefore,
1  µ µ·x  1 µ·x
ue (x) · x = − 3 +3 5 x ·x=
4π x x 2π x3
If ex is the unit vector along the radial direction (in spherical polar coor-
dinates), then the radial component of ue is

ue (x) · x 1 µ·x 1 µ cos θ

= 4
=
x 2π x 2π x3

• The stream function for a source doublet is given in equation (2.5.5) of

the book. To plot it, we write it in Cartesian coordinates. Thus,

s = x2 + y 2 + z 2
x2 + y 2
sin θ =
s2
and hence
µ x2 + y 2
ψ=
4π (x + y 2 + z 2 )3/2
2

We plot the function in the XZ plane in which y = 0. Figure 2.2 shows

the streamlines.
It was generated using the following R code.

Listing 2.2: Plotting streamlines of a source doublet

# To draw the stream lines as in the textbook ,
# i n t e r c h a n g e x and z axes . The diagram is
# drawn in XZ plane .

z <- seq ( from = -2 , to = 2 , len = 1000)

x <- seq ( from = 1e -6 , to = 2 , len = 1000)

f <- function (x , z ) {
x ^2 / ( x ^2 + z ^2)^(3 / 2)
}

psi <- outer (x , z , FUN = ’f ’)

# Values of level curves

lv <- c (1 e -6 , 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10)

# To draw the c o n t o u rs with z axis set

# horizontally , we t r a n s p o s e the psi .
Xl <- c (0.4 , 0.6)
Yl <- c (0.0 , 0.25)
contour ( t ( psi ) , levels = lv , xlim = Xl , ylim = Yl )
title ( ’ Streamlines of a source doublet ’)

A few points to note:

44
 Figure 2.2: Streamlines in an axial plane

– We don’t have to plot ue . Plotting ψ suffices to get a visual impres-

sion of the flow.
– The plot of streamlines does not give us the direction of the flow.
For an axisymmetric flow, as this one, the velocity in spherical polar
coordinates is given by equation (2.2.14) as

1 ∂ψ 1 ∂ψ
ue = es − eθ ,
s2 sin θ ∂θ s sin θ ∂s
where for sake of consistency, we have used s instead of r.
– The source doublet at x0 comprised of a point source of strength m
at x0 + δx0 /2 and a sink of strength −m at x0 − δx0 /2. Thus, the
direction of µ is along the positive z axis, shown horizontal in figure
2.2. Therefore, the direction of the velocity will be anti-clockwise
along the streamlines.
• Now let us consider a source doublet of strength µ placed at x0 + δx0 /2
and that of strength −µ placed at x0 − δx0 /2. The potential due to them
is
   
µ 1 µ 1
φe (x) = ·∇ − ·∇
4π x − x0 − δx0 /2 4π x − x0 + δx0 /2

45
 which, in Cartesian tensor notation, and some simplification gives,
 
µi ∂ 1 1
φe (x) = −
4π ∂xi s − δx0 /2 s + δx0 /2

which is,
µi ∂ s · δx0 µi ∂ sj δxj
φe (x) = =
4π ∂xi s3 4π ∂xi s3
Let νij = µi δxj . Using the fact that
 
1 s
∇ =− 3
s s
we get
νij ∂ 2
 
1
φe (x) = −
4π ∂xi ∂xj s
This expression is analogous to the potential at x due to an electric
quadrupole at x0 . The quadrupole moment tensor is νij .
• There is another way to get equations (2.5.7) and (2.5.8) of the book.
Consider an infinite line of point sources with density m. Without loss
of generality, we can make the z axis coincide it. We want to find the
velocity at a point (x, y, 0) on the XY plane. Consider a cylindrical surface
with axis along the z axis and passing through (x, y, 0). Let it have a
length L.=such that the z coordinate ranges from −L/2 to L/2. Integrate
∇· ue = ∆ over the volume of this imaginary surface. Thus,
Z Z
∇· ue ndV = ∆dV

or, since ∆ = mδ(x)δ(y),

Z Z Z L/2
ue · ndA = mδ(x)δ(y)dxdydz = mdz = mL
−L/2

Since ue is only in the radial direction, the surface integral on the left
hand side across the upper and lower surfaces is zero. (The unit vector n
on these surfaces is perpendicular to ue .) On the curved surface, ue and
n are parallel to each other and hence,

ue 2πσL = mL

or that
m 1
ue =
2π σ
or, remembering that ue is along the radial direction,
σ xex + yey xex + yey
ue = ue eσ = ur = ue p =
σ 2
x +y 2 σ
or
m x y 
ue = ex + ey
2π σ σ

46
 Since ue = ∇ φ and if the point A is a ‘reference point’ for the potential
then, Z σ Z σ
0
ue · dx = ∇ φ · dx0
A A
or
m m
ln σ − ln σA = φ(σ) − φ(σA )
2π 2π
This technique is similar to that of finding electric field due to an infinite
line of charges, distributed uniformly.

2.6 The vorticity distribution

• Unlike the velocity field, the vorticity is always solenoidal. A vortex line
is an imaginary line in the fluid which is always tangential to the vorticity
field. If dx is a small element of the vortex line, then because it is always
tangential to ω, ω ∧ dx = 0. That is,

ex (ωy dz − ωz dy) + ey (ωz dx − ωx dz) + ez (ωx dy − ωy dx) = 0

or
dx dy dz
= =
ωx ωy ωz
is the equation of the vortex lines. A surface in the fluid formed by all
vortex lines passing through an irreducible curve in the fluid is called a
vortex tube.

• A vortex tube cannot end in the interior of the fluid. For if it did, then if
C is the curve at the end of the vortex tube and A is the area enclosed by
it, Z
ω · ndA = 0
A

Let C 0 be a curve which coincides with the cross section of the vortex
tube, a little behind its end. Let A0 is the area enclosed by it. Let S be
the curved area between C 0 and C. Then, if V is the volume bound by A,
A0 and S,
Z Z Z Z
∇· ωdV = ω · ndA + ω · ndA + ω · ndA
V A S A0

Of the three integrals on the right hand side of the above equation, the
first one is zero because the vortex tube ends there, the second one is zero
because the surface is tangential to the vorticity field and the third one is
non-zero. Therefore, Z
∇· ωdV 6= 0
V
violating the fact that ω is always solenoidal.
• An immediate consequence of the above point is that vortex lines are
closed curves. (The way lines of magnetic field are closed curves.)

47
• It is also true that Z
ω · ndA
A
is the same across all cross sections of a vortex tube. It is thus a charac-
teristic of the vortex tube and is called the strength of the tube.

• Let A be an open surface inside a fluid, bound by a closed curve C. The

quantity, Z I
ω · ndA = u · dx,
C
is called the circulation round C.

• Consider a flow in which the vorticity is quite high only in the neighbor-
hood of a line. Consider a vortex tube enclosing the line. Shrink its cross
section such that as its cross section goes to zero, the integral
Z
ω · ndA
A

remains constant, say κ. Such a singularity is called a vortex line. Since

Z
ω · ndA = κ,
A

if δl is a line element along the line vortex,

Z
δl ω · ndA = κδl,
A

But δl · ndA = dV and hence,

Z
ωdV = κδl
δV

That is, for a small volume element, ωδV = κδl. Since

s ∧ ω(x0 )
Z
1
uv = − dV (x0 ),
4π s3
for a vortex line of strength κ, we have

s ∧ dl(x0 )
Z
κ
uv = − (2.6.1)
4π s3

• Consider an infinitely long line vortex of strength κ. Referring to figure

2.6.1 of the book, it is clear that s2 = σ 2 + l2 , l being measured along the
line vortex. If θ is the angle between dl and s,
σ
|s ∧ dl| = sdl sin θ = sdl sin(π − θ) = sdl = σdl
s
and hence, Z ∞
κ σ
|uv | = dl
4π −∞ (σ 2 + l2 )3/2

48
 Put l = σ tan α, so that
∞ π/2
σ 2 sec2 α
Z Z
σ
dl = dα
−∞ (σ 2 + l2 )3/2 −π/2 σ 3 sec3 α
Z π/2
1
= cos αdα
σ −π/2
2
=
σ
so that
κ
|uv | =
2πσ
Further, the direction of uv is along eφ (of cylindrical coordinate system).
Therefore, from (2.2.10) of the book,

κ ∂ψ
=−
2πσ ∂σ
or that
κ
σ=− ln σ
2π
• Since  
1 s
∇ = − 3,
s s
we have from (2.6.1),
Z  
κ 1 0
uv = ∇ ∧ dl(x )
4π s

Further, since ∇ ∧ (ψA) = ψ ∇ ∧ A + ∇ ψ ∧ A,

dl(x0 )
   
1 0
∇∧ =∇ ∧ dl(x )
s s

so that,
dl(x0 ) κdl(x0 )
Z   Z
κ
uv = ∇∧ = ∇∧
4π s 4πs
Using (4.3.1),

κdl(x0 )
Z Z  
κ 01 0
=− ∇ ∧ ndA(x )
4πs 4π s
or,
Z   Z    
κ 10 0 κ 0 1 0
uv = − ∇∧ ∇ ∧ ndA(x ) = − ∇∧ ∇ ∧ n dA(x )
4π s 4π s

Now, since n depends only on x0 ,

       
0 1 0 1 0 1
∇∧ ∇ ∧ n = n ∇· ∇ + (n · ∇)∇
s s s

49
 Further,      
01 1 2 1
∇· ∇ = − ∇· ∇ = −∇ = 0,
s s s
because 1/s is a solution of Laplace’s equation in three dimensions. Thus,
     
0 1 0 1
∇∧ ∇ ∧ n = (n · ∇)∇
s s

and hence, Z  
κ 1 0
uv = − (n · ∇)∇ dA(x0 )
4π s
Since,    
1 s 1
∇ = − 3 = −∇0
s s s
we have, Z
κ s
uv = − n·∇ dA(x0 )
4π s3
Since n depends only on x0 ,
 c s s
∇ n · 3 = n · ∇ 3 + n∧∇∧ 3
s s s
But,  
s 1
= −∇
s3 s
therefore its curl vanishes and hence,
 s s
∇ n· 3 =n·∇ 3
s s
so that, Z
κ  s
uv = − ∇ n · 3 dA(x0 )
4π s
or,
n · s
Z 
κ
uv = − ∇ dA(x0 )
4π s3
The line vortex is a closed curve and it subtends a solid angle Ω at the
point x, as shown in figure 2.3. Thus,
n · s
Z 
Ω(x) = dA(x0 )
s3
so that,
κ
uv = − ∇ Ω(x)
4π
• Equation (2.6.5) of the book gives stream function for a line vortex of
strength κ as
κ
ψ=− ln σ,
2π

50
 Figure 2.3: Solid angle subtended by a line vortex

where σ 2 = (x − x0 )2 + (y − y 0 )2 . In a vortex doublet, two line vortices

of strengths of opposite signs are close to each other. If a line vortex of
strength κ is at x0 + δx0 /2 then its stream function is
κ
ψ+ = − f (x; x0 + δx0 /2),
2π
where
f (x; x0 ) = ln σ
Similarly, a line vortex of strength −κ at x0 − δx0 has a stream function,
κ
ψ− = − f (x; x0 − δx0 /2),
2π
If u1 is associated with a vector potential B1 = (0, 0, ψ1 ) and u2 is as-
sociated with B2 = (0, 0, ψ2 ) , then it is clear that if u = u1 + u2 then
since
u = ∇ ∧ B1 + ∇ ∧ B2 = ∇ ∧ (B1 + B2 ),
the vector potential associated with u is ∇ ∧ (B1 + B2 ) = (0, 0, ψ1 + ψ2 ).
Therefore, the stream function of vortex doublet is

δx0 δx0
    
κ
ψ = ψ1 + ψ2 = − f x; x0 + − f x; x0 −
2π 2 2

Expanding f as a Taylor series and retaining the first order terms, we get

κ 0 κ δx0 · s
ψ=− δx · ∇0 f =
2π 2π σ 2
If
λ = lim
0
κδx0
δx →0

then
1 λ·s
ψ=
2π σ 2

51
• If we now want to plot the stream function, we assume that the vortex
doublet of unit strength 2π is aligned along the x-axis. That is, λ = 2πex .
If we also assume that the vortex doublet is at the origin then (x0 , y 0 ) =
(0, 0). Therefore, in Cartesian coordinates, the stream function is
1 2πx
ψ=
2π x2 + y 2
Figure 2.4 shows the stream lines.

Figure 2.4: Streamlines due to a vortex doublet

It was generated using the following R code.

Listing 2.3: Plotting streamlines of a vortex doublet

# To draw the stream lines as in the textbook ,
# i n t e r c h a n g e x and z axes . The diagram is
# drawn in XZ plane .
y <- seq ( from = -2 , to = 2 , len = 1000)
x . pos <- seq ( from = 1e -6 , to = 2 , len = 500)
x . neg <- seq ( from = -2 , to = -1e -6 , len = 500)
x <- c ( x . neg , x . pos )

f <- function (x , y ) {
( x ) / ( x ^2 + y ^2)
}

psi <- outer (x , y , FUN = ’f ’)

52
 # Values of level curves
lv <- c ( -5 , -4 , -3 , -2 , -1 , 1e -6 , 1 , 2 , 3 , 4 , 5)

# To draw the c o n t o u rs with z axis set

# horizontally , we t r a n s p o s e the psi .
Xl <- c (0.2 , 0.8)
Yl <- c (0.35 , 0.65)
contour ( psi , levels = lv , xlim = Xl , ylim = Yl )
title ( ’ Streamlines of a vortex doublet ’ ))

Since the vortex doublet consisted of vortex of strength +κ at x0 + δx0 /2

and another one of −κ at x0 − δx0 /2, the stream lines on the right have an
anti-clockwise direction and those on the left have a clockwise direction.
• Sometimes there is a considerable concentration of vorticity on a surface
in a fluid. It shows up as a large collection of line vortices, all lying on the
surface. We define the strength density of such a vortex sheet by
Z
Γ = ωdxn ,

where xn is the distance normal to the surface and the integral is over
a small range  containing the surface. We assume that as  → 0, the
integral remains constant so that Γ is a local characteristic of the vortex
sheet. Recall that the strength of a vortex tube is defined as
Z
κ = ω · ndA,

where n is a unit vector normal to the cross section of the tube over which
we are evaluating the surface integral. In particular, it is not a normal to
the vortex sheet. It is in fact, in the vortex sheet. We can write it as
ZZ Z
κ= ω · ndxdy ≈ ωdx · n (δy) = (Γ · n) δy,

or Γ · n is the strength of the vortex tube per unit width of the tube. Refer
to figure 2.5 for a diagram of the vortex sheet and the tubes it contains.
The rectangle depicts the vortex sheet and the blue circles denote the

Figure 2.5: Sheet vortex

vortex tubes. The arrow across the sheet depicts the range of integration
in the definition of Γ. The vorticity ω and the strength density of the
vortex sheet Γ are parallel to each other and perpendicular to the plane
of the diagram.
• The idea of considering  → 0 is equivalent to considering the vortex sheet
to be infinite in extent. We will use this interpretation in the next point.

53
• Let us now consider the velocity ‘due to’ a vortex sheet. Equation (2.4.11)
of the book is
s ∧ ω(x0 )
Z
1
uv = − dV (x0 )
4π s3
If most of the vorticity is concentrated in a vortex sheet, we can approxi-
mate ωdV (x0 ) by ΓdA(x0 ) so that

s ∧ Γ(x0 )
Z
1
uv = − dA(x0 )
4π s3
In the case of a single plane sheet vortex on which Γ is uniform,
Z
Γ s
uv = ∧ dA(x0 )
4π s3
Close to the sheet, the plane can be considered to be infinitely large and
yet having a uniform Γ. In such a situation, we can write s as a sum of
two vectors, one parallel to Γ and the other parallel to n. If eΓ is a unit
vector parallel to Γ then,

s = (s · eΓ )eΓ + (s · n)n

so that
Γ ∧ s = Γ ∧(s · n)n
and hence,
s·n
Z
Γ
uv = ∧ ndA(x0 )
4π s3
We can write the integral as,
Z
Γ
uv = ∧n dΩ,
4π
where Ω denotes the solid angle subtended by the sheet vortex at x. Since
we are integrating only over the “below” x, the value of the integral is half
of the total solid angle so that,
Γ 1
uv = ∧ n2π = Γ∧n
4π 2
If Γ is perpendicular to the plane of figure 2.6 and pointing outwards, the
velocity just above and just below the plane are equal and opposite to
each other.

Figure 2.6: Velocity around a sheet vortex

54
• Now consider a sheet vortex in the form of a cylinder of arbitrary cross
section. The generators of the cylinder are lines along the curved surface of
the cylinder, connecting its cross sections. Let Γ be always perpendicular
to the generators and of the same magnitude all over. Equation (2.6.8) of
the book is
s ∧ Γ(x0 )
Z
1
uv (x) = − dA(x0 )
4π s3
Consider a point x0 on the surface of the cylinder. Let dl(x0 ) be a line
element in the direction of Γ and dm(x0 ) be a line element perpendicular
to Γ. Then, Γ(x0 )dA(x0 ) = Γ(x0 )dl(x0 )dm(x0 ) = Γdl(x0 )dm(x0 ) so that
Z ∞I
Γ s ∧ dl(x0 )
uv (x) = − dm(x0 ),
4π −∞ s3

where the first integral is over m and the second one over l. Interchanging
the order of integration,
I Z ∞
Γ sdm(x0 ) 0
uv (x) = − 3
∧ dl(x ),
4π −∞ s

Write s = p + m, where p is in the cross sectional plane while m is

perpendicular to it. Further, let ep and em be the unit vectors along p
and m so that,
Z ∞ Z ∞ Z ∞
s p m
3
dm = e p 2 2 3/2
dm + em 2 2 3/2
dm
−∞ s −∞ (p + m ) −∞ (p + m )

Let the right hand side be written as ep I1 + em I2 . In I1 , put m = p tan α

so that Z π/2 2 2
1 π/2
Z
p sec α 2
I1 = 3 3
dα = cos αdα =
−π/2 p sec α p −π/2 p
In I2 , put n2 = m2 + p2 so that
Z ∞
ndn
I2 = =0
−∞ n3

and hence Z ∞
s 2 2p
3
dm = ep = 2
−∞ s p p
We now have,
p ∧ dl(x0 )
I
Γ
uv (x) = −
2π p2
• We will now argue that
p ∧ dl(x0 )
I

p2
is zero if the observation point x is outside the cylinder and 2π if it is
inside. For a point x that is inside the cylinder, we can write dl(x0 ) = pdθ
so that Z 2π
p ∧ dl(x0 )
I
= ea dθ = 2πea ,
p2 0

55
 where ea is perpendicular to both p and dl(x0 ), which means that it is
parallel to the axis of the cylinder. Now consider a point x that is outside
the cylinder. Referring to figure 2.7, let the angle between p and dl(x0 )
span from π/2 − θ (at point A) to π/2 + θ (at point C). The angle keeps
on increasing from π/2 + θ at C to 3π/2 at D and 2π + π/2 − θ = π/2 − θ
at A. Therefore,
π/2+θ π/2−θ
p ∧ dl(x0 )
I Z Z
= eb dθ + eb pdθ = 0
p2 π/2−θ π/2+θ

where eb is a unit vector in the direction of p ∧ dl(x0 ). The key to this proof
is the independence of the integrand from p. Writing dl = pdθ cancels the
second power of p in the denominator, the first power being cancelled by
p5 .

Figure 2.7: Integration when the field point is outside the cylinder

• In order to prove that the component of uv normal to the sheet vortex

is continuous across the sheet, we consider a ‘gaussian pillbox’ as shown
in figure 2.8. Let uv = up + un , where up is the component of velocity
in the plane of the sheet vortex and un is the component normal to it.
Integrating ∇· uv over the pillbox,
Z I
∇· uv dV = uv · ndA.

If the pillbox is shrunk so that its height is very small, for an incompressible
fluid, Z Z
0= uuv dA − ulv dA,
U L
where U and L are the upper and lower plane surfaces of the pillbox. Since
their areas are identical,
Z
(uuv − ulv )dA = 0,
U

or that the normal component of uv is continuous across the sheet.

5 This result is similar to the one in electrodynamics where the magnetic field outside a

solenoid is zero while inside it is parallel to the axis

56
 Figure 2.8: A gaussian pillbox straddling the sheet vortex

2.7 Irrotational and solenoidal velocity distribu-

tions
• Most fluids behave, under a wide range of flow conditions, as if they were
nearly incompressible. Further, over large parts of the velocity field, the
vorticity is indeed zero. Therefore, it is useful to study fields v that
satisfy ∇· v = 0 and ∇ ∧ v = 0. Such a fluid element does not undergo
a change in volume, nor does it have a local rotation. The second of the
above equations suggests that we can find a scalar function φ(x) such that
v = ∇ φ, which put in the first equation gives

∇2 φ = 0,

or that φ is a harmonic function. Solutions of Laplace’s equations are

smooth, with all derivatives remaining smooth throughout the domain,
except possibly at the boundary points. ( This is in contrast with the
hyperbolic equations whose solutions may have discontinuities in the form
of shocks.)
• Figure 2.7.1, to my eyes, is a multiply connected region and is identical
to fig. 2.8.1. If we have to interpret it as a singly connected region, then
we assume that there is no inner boundary for finite flows and no outer
boundary in the case of infinite flows.
• Let us consider only finite flows. Since v = ∇ φ,

∇· (φv) = φ ∇· v + v · ∇ φ = 0 + v 2 ,

so that Z Z Z
v 2 dV = ∇· (φv)dV = φv · ndA

If the normal component of the velocity on the boundary of the volume is

zero, then the integrand on the right hand side is zero so that
Z
v 2 dV = 0, (2.7.1)

which, for real values of v implies v = 0 throughout the volume. Similar

analysis works for infinite flows, where we do not have the outer boundary.
• If φ1 and φ2 are two solutions of ∇2 φ = 0 then so is φ = φ1 − φ2 . Let
v = ∇ φ, v1 = ∇ φ1 and v2 = ∇ φ2 . Equation (2.7.1), applied to v gives
v = 0 or that v1 = v2 if (v1 − v2 ) · n = 0 on the boundary.

57
• An often used mathematical idealization is that of a fluid of an infinite
extent. When we use it, the inner boundary in figure 2.7.1 comes into
play and the outer one does not. It will be shown later that one of the
conditions for having a unique solution to ∇2 φ = 0 is that the normal
component of v take a prescribed value on the inner boundary.
• Recall that the velocity of the fluid is u = ue + uv + v, where ue is
due to a given rate of expansion, uv is due to a given vorticity and v is
what is needed to satisfy boundary conditions. It is the last component
that is irrotational and solenoidal and therefore described by the velocity
potential φ. Further, the boundary conditions must be prescribed in terms
of v and not u.
• An example of a fluid of infinite extent and having the internal boundary
is when a rigid body (of size small in comparison of the extent of the fluid)
moves through a fluid otherwise at rest. If the body moves with a velocity
U then the normal component of v is
U · n − (ue + uv ) · n
Neither the body’s acceleration nor its past history determines the motion
of the fluid, only its instantaneous velocity U matters.
• Consider a two dimensional, irrotational and solenoidal field. If v = ∇ φ
and if vx = kx and vy = ly then ∇· v = 0 gives l = −k. The relation
vx = ∂φ/∂x gives φ(x, y) = kx2 /2 + f1 (y). Similarly, vy = ∂φ/∂y gives
φ(x, y) = −ky 2 /2 + f2 (x). The two expressions for φ imply that φ(x, y) =
k(x2 − y 2 )/2 + c1 , where c1 is a constant. If ψ is the stream function
corresponding to v then v = ∇ ∧ B, where B = (0, 0, ψ). That is,
∂ψ
vx =
∂y
∂ψ
vy = −
∂x
The first of the above equations gives ψ(x, y) = kxy + g1 (x) while the
second one gives ψ(x, y) = kxy + g2 (y). The two together give ψ(x, y) =
kxy + c2 , where c2 is another constant. Both φ and ψ are equations of
rectangular hyperbolae.
• If we repeat the analysis of the previous point now for a axisymmetric flow
described in a cylindrical coordinate system,
∂vx 1 ∂(σvσ )
∇· v = +
∂x σ ∂σ
If vx = kx and vσ = lσ then
1
∇· v = k + 2lσ
σ
so that ∇· v = 0 ⇒ l = −k/2. vx = ∂φ/∂x gives φ(x, σ) = kx2 /2 + f1 (σ),
while vσ = ∂φ/∂φ gives φ(x, σ) = −kσ 2 /4 + f2 (x). The two together give
σ2
 
k 2
φ(x, y) = x −
2 2

58
We can always replace the constant k/2 by another one, k. The equations
for stream function (refer to (2.2.11) in the book) are

1 ∂ψ
vx =
σ ∂σ
1 ∂ψ
vσ = −
σ ∂x
The first of these gives ψ(x, σ) = kxσ 2 /2 + g1 (x) while the second of these
gives ψ(x, σ) = kxσ 2 /2 + g2 (σ). Once again, we can replace the constant
k/2 with k and combine the two relations to get ψ(x, σ) = kxσ 2 . Figure
2.9 shows the contour plot of velocity potential and stream function. The

Figure 2.9: φ and ψ for axisymmetric flow in cylindrical coordinates.

figures were generated using the code:

Listing 2.4: Plotting velocity potential and stream function

N <- 100
x <- seq ( from = -2 , to = 2 , len = N )
sigma <- seq ( from = -2 , to = 2 , len = N )

f1 <- function (x , sigma ) { x ^2 - sigma ^2 / 2 }

f2 <- function (x , sigma ) { x * sigma ^2 }

lv <- seq ( from = -1 , to = 1 , len = 10)

phi <- outer (x , sigma , FUN = ’ f1 ’)
psi <- outer (x , sigma , FUN = ’ f2 ’)

par ( mfrow = c (1 , 2))

contour ( phi , levels = lv )
title ( ’ phi ’)
contour ( psi , levels = lv )
title ( ’ psi ’)

59
2.8 Irrotational solenoidal flow in doubly-connected
regions
• We will show that although φ is multivalued in a doubly-connected region,
v is not so. Let φ(x, n) denote the value of φ at x after going aroung the
cylinder n times. Thus, φ(x, n) = φ(x, 0) + nκ. However, ∇ φ(x, n) =
∇ φ(x, 0) + 0, which means that the velocity v does not change by going
around the cylinder.
• Equation (2.6.6) of the books says
κ
uv (x) = − ∇ Ω,
4π
where
s·n
Z
Ω(x) = dA(x0 )
s3
Since v = ∇ φ, we observe that
κ
φ(x) = − Ω,
4π
where Ω is the solid angle subtended at x by the closed line vertex. If we
go round the line vortex once, we cover a solid angle Ω + 4π so that φ gets
an additional value of −κ.
• To prove equation (2.8.5), we use another, but equivalent, definition of a
solid angle. The solid angle subtended by an area at a point is the area
of a unit sphere centerd at the point that is intersected by a cone joining
the the point to the edges of the area. Refer to figure 2.3 for a picture of
the cone. Let the line vortex be along the z axis and let it be turned into
a loop by an infinite semicircle lying in the yz plane. If the field point
is on the negative z azix, the cone collapes to a sheet and the region of
intersection with the sphere reduces to zero, making the solid angle also
zero. Let us use the negative y axis as the ‘zero’ of the angle φ. If x moves
by π/2 to reside on the positive x axis, the generators of the cone are
perpendicular to each other in the xy plane. For one of them is along the
x axis and the other is perpendicular to it, running parallel to the y axis.
If, instead, x moves only by an angle Θ, the angle between the generators
is also Θ as shown in figure 2.10. Note that the variables of integration in
Ω are the coordinates with respect to a spherical polar coordinate system
centerd at the unit sphere. Therefore, the solid angle subtended is
Z π Z Θ Z π Z Θ
Ω= sin θdθdϕ = sin θdθ dϕ = −2Θ,
0 0 0 0

so that,
κ κ
φ(x) = −(−2Θ) = Θ
4π 2π
We assumed the field point in the xy plane. There is no loss of generality
in doing so because of the infinite extent of the line vortex.

60
 Figure 2.10: Solid angle subtended by an infinite line vortex.

• In order to prove equation (2.8.9) of the book, note that v1 · v2 = ∇ φ1 ·

∇ φ2 . Now,
∇· φ1 ∇ φ2 = ∇ φ1 · ∇ φ2 + φ1 ∇2 φ2
Since φ2 is a harmonic function, ∇2 φ2 = 0 so that ∇ φ1 ·∇ φ2 = ∇· φ1 ∇ φ2
and hence, Z Z
v1 · v2 dV = ∇· φ1 ∇ φ2 dV

Using divergence theorem on the right hand side,

Z Z
v1 · v2 dV = φ1 ∇ φ2 · ndA

The bottom of page 113 showed that n · ∇ φ2 = 0, making the right hand
side of the above equation vanish.

2.8.1 Exercise
1. In equation (2.8.10), the left hand side is fixed by choice of the velocity of
the fluid. It depends only on the volume R and the external experimental
conditions. The value of the integral κ φ1 v1 · ndA is also fixed because
φ1 is single valued and v1 · n is specified by the problem. Since two out
of three terms are fixed by experiment, the third one too is fixed, that is,
independent of choice of the barrier S.
In equation (2.8.8), although
R the LHS is fixed by experiment, since φ is not
single valued, the term κ φv · ndA depends on the choice of the barrier.
Therefore, the third and remaining term of the equation too depends on
choice of the barrier S.

61
2.9 Three dimensional flow fields extending to
infinity
• Consider the expression for velocity field due to a given rate of expansion,
Z ∞ Z 2π Z π
1 s 2
ue = 3
∆(x0 )r0 sin θ0 dr0 dθ0 dϕ0
4π 0 0 0 s

Divide the domain of integral into an inner spherical region of area αr and
the one outside it. The right hand side than then be written as a sum of
two integrals, I1 and I2 of which
Z ∞ Z 2π Z π
1 s 2
I2 = 3
∆(x0 )r0 sin θ0 dr0 dθ0 dϕ0
4π αr 0 0 s
−n
Let ∆(x0 ) = ar0 . In the region of integration, s ≈ r, where r = |x|.
Therefore,
Z ∞ Z 2π Z π
1 r −n 0 2
I2 = ar0 r sin θ0 dr0 dθ0 dϕ0
4π r3 αr 0 0

or
er a
I2 = = er O(r−n+1 )
r (αr)−3+n
2

Therefore, when the field point x is very far away from the source point
x0 , we can approximate ue (x) as
Z αr Z 2π Z π
1 x 2
ue = ∆(x0 )r0 sin θ0 dr0 dθ0 dϕ0
4π 0 0 0 x3

Since  
1 x
∇ =− 3
r x
where r = |x|, we have
Z αr Z 2π Z π  
1 1 2
ue = − ∇ ∆(x0 )r0 sin θ0 dr0 dθ0 dϕ0
4π 0 0 0 r
or   Z αr Z 2π Z π 
1 1 2
ue = − ∇ ∆(x0 )r0 sin θ0 dr0 dθ0 dϕ0
4π r 0 0 0

or, writing the integral more succinctly,

  Z 
1 1 0 0
ue = − ∇ ∆(x )dV (x )
4π r

If the integral ∫ ∆(x0 )dV (x0 ) vanishes, we go to the next term in the Taylor
expansion of s/s3 and write ue (x) in terms of the source dipole. Refer
to point 4.4 for the ‘multipole’ expansion of electrostatic field, which is
analogous to ue .

62
• Velocity field due to a given vorticity distribution is

s ∧ ω(x0 )
Z
1
uv (x) = − dV (x0 )
4π s3
Proceeding in similar was as we did for a given rate of expansion, for a
localized vorticity distribution, we get
  Z 
1 1 0 0
uv = ∇ ∧ ω(x )dV (x )
4π r
Z   
1 0 0 1
= − ω(x )dV (x ) ∧ ∇
4π r

Consider ∇· (xi ω) = ∇ xi · ω + xi ∇· ω = ei · ω = ωi . Therefore,

Z Z
ω(x0 )dV (x0 ) = ∇· (x0i ω(x0 ))dV (x0 )

Using divergence theorem on the right hand side,

Z Z
ω(x )dV (x ) = xi ω(x0 ) · ndA
0 0

If we choose α such that on the surface of the sphere of radius αr, the
vorticity is very small, then the right hand side of the above equation is
zero.

• We now prove the identity,

A ∧(B · ∇ f ) = ∇ ∧ (A(B · f )),

where A and B are constant vectors while f is an irrotational vector field.

Starting from the right hand side,

∇ ∧ (A(B · f )) = (∇ ∧ A)(B · f ) + A ∧ ∇ (B · f )

which follows immediately if we write φ(x) = B · f . Since, A is a constant

vector,
∇ ∧ (A(B · f )) = A ∧ ∇ (B · f )
Now,
∇ (B · f ) = B · ∇ f + f · ∇ B + B ∧ ∇ ∧ f + f ∧ ∇ ∧ B
Since B is a constant vector and f is an irrotational field,

∇ (B · f ) = B · ∇ f

Therefore,
∇ ∧ (A(B · f )) = A ∧ (B · ∇ f ) (2.9.1)

• We go back to the velocity field due to a given vorticity distribution

s ∧ ω(x0 )
Z
1
uv (x) = − dV (x0 )
4π s3

63
 and write it as
Z  
1 1 0 0
uv (x) = ∇ ∧ ω(x )dV (x )
4π s

We know that  
1 1 1
= + x0 · ∇
s r s
When r = |x| is very large as compared to r0 = |x0 |, we can approximate
s as r so that  
1 1 1
= + x0 · ∇
s r r
We saw previously that approximating s−1 as r−1 led to vanishing velocity.
We therefore consider the second term to get
Z   
1 1
uv (x) = ∇ x0 · ∇ 0 0
∧ ω(x )dV (x )
4π r

Now since x0 is a constant in a gradient operation,

        
1 1 1
∇ x0 · ∇ = x0 · ∇ ∇ + x0 ∧ ∇ ∧ ∇
r r r

Since curl of a gradient is always zero,

     
0 1 0 1
∇ x ·∇ =x ·∇ ∇
r r

so that Z   
1 1
uv (x) = x0 · ∇ ∇ 0 0
∧ ω(x )dV (x )
4π r
or Z    
1 1
uv (x) = − ω(x0 ) ∧ x0 · ∇ ∇ dV (x0 )
4π r
From (2.9.1),
Z    
1 1
uv (x) = − ∇ ∧ ω(x0 ) x0 · ∇ dV (x0 )
4π r

Since curl is with respect to unprimed coordinates.

Z   
1 1
uv (x) = − ∇ ∧ ω(x0 ) x0 · ∇ dV (x0 ) (2.9.2)
4π r

• We will now prove that

Z
(xi ωj + xj ωi )dV = 0

and if ω decays faster than r−4 , the right hand side is zero. Consider

∇· (xi xj ω) = ∇ (xi xj ) · ω + xi xj ∇· ω = ∇ (xi xj ) · ω

64
 Now, ∇ (xi xj ) = xi ∇ xj + xj ∇ xi = xi δkj ek + xj δik ek = xi ej + xj ei so
that
∇ (xi xj ) · ω = (xi ej + xj ei ) · ω = xi ωj + xj ωi
and hence Z Z
∇· (xi xj ω)dV = (xi ωj + xj ωi )dV

The left hand side can be written as

Z
(xi xj ω) · ndA

If ω decays faster than r−4 then the volume over which we integrated can
be made large enough so that ω is almost zero on its surface making
Z
(xi ωj + xj ωi )dV = 0 (2.9.3)

Writing in Gibbs vector form,

Z
(xω + ωx)dV = 0 (2.9.4)

• An immediate consequence of (2.9.4) is that if f is a vector field

Z
(xω · f + ωx · f )dV = 0 (2.9.5)

so that
Z Z Z
1
− ωx · f dV = − ωx · f dV + (xω · f + ωx · f )dV
2
or Z Z
1
− ωx · f dV = (xω · f − ωx · f )dV
2
If f = ∇ (1/r),
Z   Z    
1 1 1 1
− ωx · ∇ dV = (xω · ∇ − ωx · ∇ dV
r 2 r r

Since
     
1 0 0 0 1 0 0 1
∇ ∧ (ω(x ) ∧ x ) = ω(x ) ∇ ·x −x ∇ · ω(x0 )
r r r

we have
     
0 1 0 0 1 0 1 0 0
x ∇ · ω(x ) − ω(x ) ∇ · x = −∇ ∧ (ω(x ) ∧ x )
r r r

so that
Z   Z  
1 1 1
− ω(x0 )x0 · ∇ dV = − ∇ 0 0
∧ (ω(x ) ∧ x ) dV
r 2 r

65
 or Z   Z  
1 1 1
− ω(x0 )x0 · ∇ dV = ∇ 0 0
∧ (x ∧ ω(x )) dV
r 2 r
From (2.9.2),
Z  
1 1 0 0 0
uv (x) = ∇∧ ∇ ∧ (x ∧ ω(x )) dV (x )
8π r
or    Z 
1 1 0 0 0
uv (x) = ∇∧ ∇ ∧ (x ∧ ω(x )) dV (x ) (2.9.6)
8π r
If A is a constant vector and f is a function of x then ∇ (A · f ) = A · ∇ f +
A ∧ ∇ ∧ f and ∇ ∧ (f ∧ A) = −A ∇· f + A · ∇ f . Using the first equation in
the second, we get
∇ ∧ (f ∧ A) = −A ∇· f + ∇ (A · f ) − A ∧ ∇ ∧ f
If f = ∇ (1/r), then its curl and divergence are zero so that
       
1 1
∇∧ ∇ ∧A = ∇ ∇ ·A
r r
Using this in (2.9.6), we get
   Z 
1 1 0 0 0
uv (x) = ∇ ∇ · (x ∧ ω(x )) dV (x ) (2.9.7)
8π r
The book uses an approximation sign instead of equality because this
relation is true only very far away from the source.
• In the section on the behaviour of φ at large distances, the analysis be-
gins with Green’s theorem whose requirements are that the functions F
and G be single-valued, finite and continuous throughout the volume of
integration. The same restrictions are placed on derivatives of F and G.
These requirements force us to exclude the point P from the region of
integration by enclosing it in a sphere of radius .
• In order to get (2.9.9) we evaluate the limit
−1
1 ∂φ(x0 ) 
Z  
0 ∂s
L = − lim φ(x ) −  2 dΩ(x0 )
→0 ∂s s ∂s s=

The integrand is simplified as

φ(x0 ) 1 ∂φ(x0 ) ∂φ(x0 )
 
1
− − =− 2 φ(x0 ) + s
s2 s ∂s s ∂s
Noting that the position vector of P is x we replace the primed variable
with the unprimed one. Evaluated at s =  it is
 
1 ∂φ(x) φ(x + )
− 2 φ(x) +  =− ,
 ∂s 2
where  is the radius vector of the sphere around P . We further get,
Z
φ(x + ) 2
L = + lim  dΩ(x) = 4πφ(x)
→0 2

66
• After accounting for the above term, equation (2.9.8) of the book becomes
R 
R (F ∇ G − G ∇ F ) · n2 dA2 − 
Z
(F ∇ G − G ∇ F ) · n1 dA1 + = (F ∇2 G − G∇2 F )dV
+4πφ(x)


If F (x0 ) = φ(x0 ) and G(x0 ) = s−1 then the right hand side of the above
equation is zero and
Z Z
4πφ(x) = (F ∇0 G − G ∇0 F ) · n1 dA1 − (F ∇0 G − G ∇0 F ) · n2 dA2 ,

where ∇0 denotes the gradient with respect to primed coordinates. Let T2

denote the second term.
Z
T2 = (F ∇0 G − G ∇0 F ) · n2 dA2
Z      
1 1
= φ(x0 ) ∇0 − ∇0 φ(x0 ) · n2 dA2
s s

Since A2 is the surface of a sphere of radius R,

Z Z
1 1
T2 = − 2 0
φ(x )dA2 − ∇0 φ(x0 ) · n2 dA2
R R
Since ∇· v = 0 everywhere in V ,
Z
∇· vdV = 0

or Z Z
v · n2 dA2 − v · n1 dA1 = 0

Let Z Z
∇0 φ(x0 ) · n2 dA2 = ∇0 φ(x0 ) · n1 dA1 = m,

say so that
Z
1 m m
T2 = − φ(x0 )dA2 − = 4πhφ(x, R)i − ,
R2 R R
where h·i denotes the average. Thus,
Z
m
4πφ(x) = (F ∇0 G − G ∇0 F ) · n1 dA1 + 4πhφ(x, R)i −
R
or
Z      
m 1 1 1
φ(x) = hφ(x, R)i − + φ(x0 ) ∇0 − ∇0 φ(x0 ) · n1 dA1
4πR 4π s s
Note that the gradients in the integrand are all with respect to primed
coordinates. If we want them to be in terms of un-primed coordinates, we
get an overall negative sign to the integral.
Z      
m 1 1 1
φ(x) = hφ(x, R)i − − φ(x0 ) ∇ − ∇ φ(x0 ) · n1 dA1
4πR 4π s s
(2.9.8)

67
• We can write the equation for flux of φ as

∂φ(x0 )
Z Z  
m = ∇0 φ(x0 ) · n2 dA2 = R2 dΩ(x0 )
∂s s=R

Pulling out the differential operator (reverse of integration under the dif-
ferential sign), Z
2 ∂
m=R (φ(x0 ))s=R dΩ(x0 )
∂R
or, Z
mdR
= ∂ (φ(x0 ))s=R dΩ(x0 )
R2
or Z
m
4πC − = (φ(x0 ))s=R dΩ(x0 ),
R
where C is a constant of integration. Using the definition of hφ(x, R)i, we
see that
m
hφ(x, R)i = C −
4πR
• Since R is a fixed constant, ∇ hφ(x, R)i = ∇ C so that
 Z  Z
1 0 1
∇C = ∇ φ(x ) · ndA 2 = ∇ x0 · ndA2
4πR2 4πR2
or Z Z
1 1
∇C = − ∇0 x0 · ndA2 = − v · ndA2
4πR2 4πR2
Since v is zero on the outer boundary, we see that C = 0 or that C is
independent of R.
• Since
m
C = hφ(x, R)i +
4πR
we can write (2.9.8) as
Z      
1 1 1
φ(x) = C − φ(x0 ) ∇ − ∇ φ(x0 ) · n1 dA1
4π s s

This does not appear to be a very encouraging equation because φ appears

on the right hand side as well. It appears as ‘prescribed value’ of φ on the
inner boundary. The book mentions that it can be converted in principle
into the prescribed value of ∇ φ · n on the surface, although it does not
tell how so.

• We observed that Z Z
v · vdV = ∇· (φv)dV

Since v is a solenoidal field, v · v is also equal to ∇· ((φ − C)v). Applying

this relation in the present context,
Z Z Z
v · vdV = lim ((φ − C)v) · n2 dA2 − ((φ − C)v) · n1 dA1
R→∞

68
 The limit is applicable only to the first term because the inner boundary is
fixed while the outer, spherical one is assumed to be at an infinite distance.
Since φ(x) → C in the limit R → ∞, the first term goes to zero and
Z Z
v · vdV = − ((φ − C)v) · n1 dA1
or
Z Z Z Z
v · vdV = − φv · ndA1 + C v · n1 dA1 = − φv · ndA1 + Cm,

where m is the flux of φ.

• We previously observed that (see bottom of page 87 of the book) that v
is determined from the prescribed boundary conditions. We will find v
under the condition v · n = U · n, where U is the velocity of a rigid body
in a translational motion in the fluid, which is at rest at infinity. Thus,
the mathematical statement of the problem is - solve ∇2 φ = 0 subject to
the boundary conditions
1. φ(x) → a constant as r = |x| → ∞. We can choose the constant to
be zero.
2. n · ∇ φ = n · U
We try a solution of the form φ(x) = U · Φ(x), where Φ is independent
of U. Since Φ is determined by the boundary condition on the surface of
the rigid body, its value depends not on x but on its position vector with
respect to an arbitrarily chosen material point, say x0 , of the body. Thus,
φ(x) = U · Φ(x − x0 )
For a spherically symmetric body with its center instantaneously at the
origin, x0 = 0. Further,
∇2 φ = ∇2 (Ui Φi ) = Ui ∇2 Φi
Since Ui is, in general, not zero, ∇2 φ = 0 implies ∇2 Φi = 0 or ∇2 Φ = 0.
A solution of Laplace’s equation which is also a component of a vector is
 
∂ 1
∂xi r
Therefore,  
1
Φ = α∇ ,
r
where α is an arbitrary constant. Therefore,
 
1 U·x
φ(x) = αU · ∇ = −α 3
r r
and hence,
v = ∇φ
  
1
= −α ∇ U · ∇
r
    
1 1
= −α(U · ∇) ∇ − αU ∧ ∇ ∧ ∇
r r

69
 Since curl of a gradient is always zero,
 
1
v = −α(U · ∇) ∇
r

2.10 Two dimensional flow fields extending to

infinity
• A two dimensional motion is the one whose velocity field is of the form
u(x, y) = (u, v, 0) or equivalent. A two dimensional flow field extending to
infinity is the one which extends to infinity in x and y (or any other pair)
of directions. It is either bounded in the z (or the remaining) direction
or it extends to infinity without motion diminishing as one travels along
the z direction. In such a flow field, the assumption made in the previous
section that the velocity on the outer sphere is zero does not hold good.
That is why it needs a separate treatment.
• The construct of a large sphere, centered at the field point and including
all internal boundaries is used only to examine the behaviour of φ. There-
fore, all formulae in section 2.9 of the book, up to equation (2.9.7), are
applicable to two dimensional flow fields.

• It is possible to use equation (2.9.8) of the book for the plane by replacing
dV by dA and dA by dl because the divergence theorem can be applied
to plane as well. Refer to sub section 4.7 for a proof.
• We saw in section 4.6.1 that ln |x − x0 | is a solution of Laplace’s equation
in two dimensions. That is why we choose F (x0 ) = φ(x0 ) and G(x0 ) = ln s
in equation(2.9.8). Doing so, we get
Z Z
∇ (φ(x ) ln s) · dl2 − ∇0 (φ(x0 ) ln s) · dl01 =
0 0 0

Z
2 2
(φ(x0 )∇0 ln s − ln s∇0 φ(x0 ))dA0 ,

where the subscript 1 to ∇ indicates differentiation with respect to the

primed coordinates. The function ln s blows up at the field point x. There-
fore, we exclude it by enclosing it in a small circle of radius  centerd at x.
The exclusion appears on the left hand side in the form of an additional
term
∂φ(x0 ) 
Z  
0 ∂(ln s)
L = − lim φ(x ) − ln s  dθ
→0 ∂s ∂s s=

Interchanging the order of integration and limit the integrand is,

φ(x0 ) ∂φ(x0 ) 
 
lim − ln   = φ(x0 ),
 ∂s s=

→0

because
lim  ln  = 0,
→0

70
 by l’Hôpital’s rule. Therefore, L = −2πφ(x0 ). Since φ and ln s are har-
monic functions, the right hand side of the 2d analog of (2.9.8) is
Z Z
∇0 (φ(x0 ) ln s) · dl02 − ∇0 (φ(x0 ) ln s) · dl01 − 2πφ(x0 ) = 0

or Z Z
1 1 0
φ(x) = 0 0
∇ (φ(x ) ln s) · dl02 − ∇ (φ(x0 ) ln s) · dl01
2π 2π
Since the first integral on the right hand side is taken over a circle of radius
R centerd at x,
Z Z Z
1 ln R 0 1
φ(x) = 0 0
φ(x )dl2 + 0 0
∇ φ(x ) · dl2 − ∇0 (φ(x0 ) ln s) · dl01
2πR 2π 2π
(2.10.1)
Now, Z
1
φ(x0 )dl20 = hφi,
2πR
the mean value of φ over the outer boundary while
Z Z
− ∇0 φ(x0 ) · dl02 = ∇ φ(x0 ) · dl02 = m,

is the flux through it, so that (2.10.1) becomes,

Z
m 1
φ(x) = hφi − ln R − ∇0 (φ(x0 ) ln s) · dl01 (2.10.2)
2π 2π

• The analog of the flux relation (2.9.10) is

Z
m = ∇0 φ(x0 ) · dl02

Over a circle of radius R centerd at x,

Z Z
∂φ  ∂ 
m=R dθ(x0 ) = R φ(x0 ) dθ(x0 )

∂s s=R ∂R

s=R

or Z 
m ln R = φ(x0 ) dθ(x0 ) − 2πC,

s=R

where C is a constant of integration. Thus,

m
ln R = hφi − C
2π
so that equation (2.10.2) becomes
Z
1
φ(x) = C − ∇0 (φ(x0 ) ln s) · dl01
2π
In terms of derivative with respect to unprimed coordinates,
Z
1
φ(x) = C + ∇ (φ(x0 ) ln s) · dl01
2π

71
 or Z
1
φ(x) = C + (φ(x0 ) ∇ ln s + ln s ∇ φ(x0 )) · dl01 (2.10.3)
2π
or
φ(x0 )
Z  
1
φ(x) − C = + ln s ∇ φ(x ) · dl01
0
2π s
or
Z 
m 1 s  m
φ(x) − C − ln r = φ(x0 ) + ln s ∇ φ(x0 ) · dl01 − ln r
2π 2π s 2π
Using the fact that Z
m= ∇0 φ(x0 ) · dl02

on the right hand side we get

Z 
m 1 s s 
φ(x) − C − ln r = φ(x0 ) + ln ∇ φ(x0 ) · dl01
2π 2π s r
Unlike the three dimensional case, as r → ∞, it is not φ but
m
φ(x) − ln r
2π
that tends to C. We need the term m ln r/(2π) because it creates the term
ln(s/r), whose argument to 1 as r → ∞.
• We can reconcile the behaviour in two and three dimensions if we state
that it is the difference between the velocity potential and the source term
that tends to a constant as r → ∞. In three dimensions, the source term
being of O(r−1 ) itself vanishes at infinity. On the other hand, in two
dimensions, the source term rises logarithmically at infinity.
• We first observe that
h m i h m i
v−∇ ln r · v − ∇ ln r =
nh  2π
m i h  m2π io
∇· φ − C − ln r v−∇ ln r
2π 2π
This is because v is a solenoidal field and ln r is a harmonic function.
Therefore,
Z h m i h m i
v−∇ ln r · v − ∇ ln r dA =
2π 2π
Z nh m i h m io
∇· φ − C − ln r v−∇ ln r dA
2π 2π
Using divergence theorem in the plane (refer to 4.7) on the right hand side
Z h m i h m i
v−∇ ln r · v − ∇ ln r dA =
2π 2π
Z nh m i h m io
φ−C − ln r v−∇ ln r · n2 dl−
2π 2π
Z nh m i h m io
φ−C − ln r v−∇ ln r · n1 dl
2π 2π

72
 Choose the outer boundary as large as possible by making the radius of
the outer circle tend to infinity. In this limit the integrand tends to zero
and hence,
Z h m i h m i
v−∇ ln r · v − ∇ ln r dA =
2π 2π
Z nh m i h m io
− φ−C − ln r v−∇ ln r · n1 dl
2π 2π
If
m 
Φ = φ− ln r
2π
m 
V = v−∇ ln r
2π
then the previous equation becomes,
Z Z
V · VdA = − (Φ − C)V · n1 dl

Now let, if possible, Φ and Φ∗ be two solutions of Laplace’s equation in

two dimensions with identical boundary conditions. Then Φ − Φ∗ too will
be a solution and
Z Z
(V − V∗ ) · (V − V∗ )dA = − ((Φ − Φ∗ ) − (C − C ∗ )) (V − V∗ ) · n1 dl

By virtue of boundary conditions, ∇ φ · n = v · n is identical on the

boundary for both solutions. Since the factor m ln r/(2π) is same for both
solutions,
V · n = V∗ · n
on the boundary. Therefore the right hand side is zero making
Z
(V − V∗ ) · (V − V∗ )dA = 0,

which is possible only if V = V∗ . Thus, in two dimensions the unique

solution of Laplace equation is
m 
φ− ln r
2π

2.11 Exercises
1. Consider a material line element joining points P and Q. Let r be the
length of the material line element. If x0 is the position vector of P then
that of Q is x0 + rer , where er is a unit vector parallel to the line element,
at a certain instance of time. Let, after a time δt, P get to x0 + X so that
Q goes to (x0 + rer ) + rer · ∇ X, where we have ignored terms of higher
order in rer . If the line element was along rer initially, it is now along
qeq = r(er + er · ∇ X). Thus,

q 2 = r2 (1 + 2er · ∇ X · er ),

73
 where we have ignored terms higher than second order in er . Thus,

q2
= 1 + 2er · ∇ X · er
r2
or
q 1/2
= (1 + 2er · ∇ X · er ) = 1 + er · ∇ X · er
r
or
q q−r
−1= = er · ∇ X · er
r r
The quantity (q − r)/r is the strain of the element, λ. Therefore, the rate
of change of strain is,
dλ
= er · ∇ U · er ,
dt
where U is the velocity of Q relative to P . We can as well write it as
dλ r · ∇U · r
=
dt r2
The quantity r · ∇ U · r is the rate-of-strain ellipsoid centerd at P . The
point Q lies on it. If one imagines a sphere centered at P and of radius
r then Q lies on it in the unstrained state. As time goes by, the point Q
finds itself on the ellipsoid described by r · ∇ U · r.
2. Given that
ω(x0 ) n ∧ u(x0 )
Z Z
1 1
Bv (x) = dV (x0 ) − dA(x0 )
4π s 4π s
Therefore,

ω(x0 ) n ∧ u(x0 )
Z Z
1 1
∇ ∧ Bv (x) = ∇∧ dV (x0 ) − ∇∧ dA(x0 )
4π s 4π s
Using the fact that  
1 s
∇ = − 3,
s s
we get

u(x) = u1 (x) + u2 (x)

s ∧ ω(x0 ) s ∧(n ∧ u(x0 ))
Z Z
1 0 1
= − dV (x ) + dA(x0 )
4π s3 4π s3

We will now show that ∇ ∧ u2 (x) = 0 that is u2 (x) does not contribute to
the vorticity of the fluid.

s ∧(n ∧ u(x0 ))
Z
1
∇ ∧ u2 (x) = ∇∧ dA(x0 )
4π s3
(nu(x0 ) − u(x0 )n) · s
Z
1
= ∇∧ dA(x0 )
4π s3
Z
1 s
= (nu(x0 ) − u(x0 )n) · ∇ ∧ 3 dA(x0 )
4π s

74
 Now
s ∇∧s 1
∇∧ = + s∧∇ 3
s3 s3 s
s
= 0 − s∧ 5
s
= 0
Thus ∇ ∧ u2 (x) = 0. We will now show that ∇ ∧ u1 (x) = ω(x).
s ∧ ω(x0 )
Z
1
∇ ∧ u1 (x) = − ∇∧ dV (x0 )
4π s3
Now
s ∧ ω(x0 ) 0 s 0
s
∇∧ = −ω(x ) ∇· − ω(x ) · ∇ ,
s3 s3 s3
where we have used the fact6 that
s
∇· 3 = 4πδ(x − x0 )
s

s ∧ ω(x0 )
 
0 0 ∇s0 1
∇∧ = −ω(x )4πδ(x − x ) − ω(x ) · + s∇ 3
s3 s3 s
e e ss 
s s
= −4πω(x0 )δ(x − x0 ) − ω(x0 ) · −
s3 s5
Since s = ses , the last factor is 0. Therefore,
Z
∇ ∧ u1 (x) = ω(x0 )δ(x − x0 )dV (x0 ) = ω(x)

Thus
∇ ∧ ∇ ∧ Bv (x) = ∇ ∧ u1 (x) + ∇ ∧ u2 (x) = ω(x)

3. Green’s theorem is
Z Z
(f ∇ g − g ∇ f ) · ndA(x) = (f ∇2 g − g∇2 f )dV (x)

We will now consider the three cases:

(i) Choose
1
f =
4πs
g = φ(x)
in the statement of Green’s theorem. Now,
Z  
∇ φ(x)
Z
1
(f ∇ g) · ndA(x) = · ndA(x)
4π s
Z Z  
1 1
(g ∇ f ) · ndA(x) = φ(x)n · ∇ dA(x)
4π s
Z  
1 1
= µ·∇ dA(x),
4π s
6 See chapter 1 of Griffith’s Classical Electrodynamics

75
 where we put µ = φ(x)n. The first term is the velocity induced
by source distribution of ∇ φ(x) over the surface (refer to equation
(2.9.20) of the book) while the second one is due to velocity induced
due to source doublet distribution of µ (refer to equation (2.5.3) of
the book). The integrand of right hand side of Green’s theorem is
 
1 2 1
∇ φ − φ∇2
4πs 4πs

Since φ(x) satisfies Laplace’s equation, the integrand is just the sec-
ond of the above terms. Therefore, the right hand side is
Z   Z 
1 s 
− φ(x)∇2 dV (x) = φ(x) ∇· dV (x)
4πs 4πs3
Z
= φ(x)δ(x − x)dV (x)

= φ(x)

Thus φ(x) is due to a source distribution of ∇φ(x) over the surface

and a source doublet distribution of φ(x)n.
(ii) In this case, we apply Green’s theorem separately to the region, Ro ,
outside the boundary and Ri , the one inside it. In Ro , choose
1
f =
4πs
g = φ∗ (x)

while in Ri choose
1
f =
4πs
g = φ(x)

Green’s theorem applied in Ri gives,

Z Z   Z  
1 1 1 1 1 1
∇ φ · ni dA − φ∇ · ni dA = − φ∇2 dV,
4π s 4π s 4π Ri s
(2.11.1)
where we used the fact that ∇2 φ = 0. Doing the same in Ro gives,
Z Z  
1 1 ∗ 1 ∗ 2 1
∇ φ · no dA− = − φ ∇ dV
4π s 4π Ro s
Z  
1 1
φ∗ ∇ · no dA
4π s

Since no = −ni , we have

Z Z  
1 1 1 1
− ∇ φ∗ · ni dA+ = − φ ∗ ∇2 dV (2.11.2)
4π s 4π Ro s
Z  
1 ∗ 1
φ ∇ · ni dA
4π s

76
 Adding (2.11.1) and (2.11.2),
Z Z  
1 1 1 1
∇ (φ − φ∗ ) · ni dA− = − φ∇2 dV
4π s 4π Ri s
Z   Z  
1 1 1 1
(φ − φ∗ ) ∇ · ni dA − φ ∗ ∇2 dV
4π s 4π Ro s

Since φ = φ∗ on the boundary,

∇ (φ − φ∗ ) · ni
 
−1
Z Z
1 ∗ 2 1
dA = φ ∇ dV +
4π s 4π Ro s
 
−1
Z
1
φ∇2 dV
4π Ri s

We cannot combine the integrals on the right hand side. We there-

fore, interpret this equation separately in the two regions,
Z Z  
1 1 ∗ 1 2 1
∇ (φ − φ ) · ni dA = − φ∇ dV
4π s 4π Ri s
Z Z  
1 1 1 1
∇ (φ − φ∗ ) · ni dA = − φ ∗ ∇2 dV
4π s 4π Ro s

Noting that  
1
∇2 = 4πδ(x − x0 )
s
we get,
Z
1 1
∇ (φ − φ∗ ) · ni dA = −φ(x)
4π s
Z
1 1
∇ (φ − φ∗ ) · ni dA = −φ∗ (x)
4π s
Thus, in both regions, the potential can be interpreted as due to
sources of strength ∇ (φ − φ∗ ) · ni .
(iii) We once again apply Green’s theorem separately to the region, Ro ,
outside the boundary and Ri , the one inside it. In Ro , choose
1
f =
4πs
g = φ∗ (x)

while in Ri choose
1
f =
4πs
g = φ(x)

Green’s theorem applied in Ri gives,

Z Z   Z  
1 1 1 1 1 1
∇ φ · ni dA − φ∇ · ni dA = − φ∇2 dV,
4π s 4π s 4π Ri s
(2.11.3)

77
where we used the fact that ∇2 φ = 0. Doing the same in Ro gives,
Z Z  
1 1 1 1
∇ φ∗ · no dA− = − φ ∗ ∇2 dV
4π s 4π Ro s
Z  
1 1
φ∗ ∇ · no dA
4π s

Since no = −ni , we have

Z Z  
1 1 1 1
− ∇ φ∗ · ni dA+ = − φ ∗ ∇2 dV (2.11.4)
4π s 4π Ro s
Z  
1 1
φ∗ ∇ · ni dA
4π s

Adding (2.11.3) and (2.11.4),

Z Z  
1 1 1 1
∇ (φ − φ∗ ) · ni dA− = − φ∇2 dV
4π s 4π Ri s
Z   Z  
1 1 1 1
(φ − φ∗ ) ∇ · ni dA − φ ∗ ∇2 dV
4π s 4π Ro s

Since ∇ φ · ni = ∇ φ∗ · ni on the boundary,

   
−1
Z Z
1 1 1
(φ − φ∗ ) ∇ · ni dA = φ ∗ ∇2 dV
4π s 4π Ro s
 
−1
Z
1
+ φ∇2 dV
4π Ri s

We cannot combine the integrals on the right hand side. We there-

fore, interpret this equation separately in the two regions,
Z Z  
1 1 1 1
(φ − φ∗ ) ∇ · ni dA = − φ∇2 dV
4π s 4π Ri s
Z Z  
1 1 1 1
(φ − φ∗ ) ∇ · ni dA = − φ ∗ ∇2 dV
4π s 4π Ro s

Noting that  
1
∇2 = 4πδ(x − x0 )
s
we get,
Z
1 1
(φ − φ∗ ) ni · ∇ dA = −φ(x)
4π s
Z
1 1
(φ − φ∗ ) ni · ∇ dA = −φ∗ (x)
4π s
Thus, potential in either regions can be considered to be due to source
doublets of strength (φ − φ∗ ) ni on the surface.

78
4. The potential due to a line vortex of strength κ is given by (2.8.4) to be
κ
φ(x) = − Ω,
4π
where Ω(x) is the solid angle subtended by the curve s at a point x is
given by
s·n
Z
Ω= dA(x0 )
s3
Therefore
s·n
Z
κ
φ(x) = − dA(x0 )
4π s3
Z  
1 1
= κn · ∇ dA(x0 )
4π s
Z  
1 1
= µ·∇ dA(x0 )
4π s
The last equation suggests that φ(x) can be interpreted as potential due a
distribution of source doublets of strength µ = κn over the surface whose
boundary coincides with that of the vortex tube. We will now show that
the velocity corresponding to this potential is solenoidal and irrotational.
Let u = ∇ φ, so that
Z   
1 1
u= ∇ µ·∇ dA(x0 )
4π s
Therefore,
Z   
1 1
∇∧u = ∇∧∇ µ · ∇ dA(x0 ) = 0
4π s
making u an irrotational field. Further,
Z   
1 1
∇· u = ∇· ∇ µ · ∇ dA(x0 )
4π s
Z   
1 1
= 2
∇ µ·∇ dA(x0 )
4π s
Z   
1 1
= µ·∇ ∇ 2
dA(x0 )
4π s
Z   
1 1
= µ · ∇∇ 2
dA(x0 )
4π s
Z
= µ · ∇ δ(x − x0 )dA(x0 )
Z
= − δ(x − x0 ) ∇· µdA(x0 )
Z
= − δ(x − x0 ) · (0)dA(x0 ) = 0,

making u a solenoidal field. In getting from the fifth to the sixth step, we
used the identity
Z Z
f (x0 ) · ∇ δ(x − x0 )dA(x0 ) = − δ(x − x0 ) ∇· f (x0 )dA(x0 ) (2.11.5)

79
We can prove it using integration by parts. In Cartesian tensor form, the
left hand side is
Z Z
0 ∂ ∂fi

fi (x ) δ(x − x )dA = fi (x )δ(x − x ) − δ(x − x0 )
0 0 0 
dA
∂xi C ∂xi
where C denotes the boundary of the area of integration. Since the delta
function is zero at all points other than x0 ,
Z Z
0 ∂ ∂fi
fi (x ) δ(x − x )dA = − δ(x − x0 )
0
dA
∂xi ∂xi

which, in Gibbs notation gives (2.11.5). We still have to prove:

• A sheet vortex is equivalent to a distribution of surce doublets over

and normal to the surface coinciding with the sheet, provided that
the closed vortex lines are reducible on the sheet. To that end, once
again consider figure 2.5. In that figure, the blue circles depict the
vortex tubes. Since vortex tubes are closed, each one of them has a
stream of fluid getting in the plane and another on getting out of the
plane. A cross section of the vortex tube can thus be imagined to
be an assembly of a source and a sink. Further, this assembly of a
source and a sink is such that one of them is top of the other. (They
are not side-by-side or in any other orientation.) Thus, the source
doublet associated with the assembly is normal to the sheet.
• Any irrotational, solenoidal motion can be regarded as due to a cer-
tain sheet vortex coinciding with the boundary of the region of mo-
tion. To that end, we begin with (4.9.1).
Z Z
φ1 ∇ φ2 · ndA = φ2 ∇ φ1 · ndA

Choose φ2 = |x − x0 |−1 = s−1 . Further, let the integration be over a

surface containing the field point P with position vector x. Thus,
Z   Z  
1 1
φ1 ∇ · ndA = ∇ φ1 · ndA
|x − x0 | |x − x0 |

Since the field point is on the surface of integration, the integral on

the left hand side diverges. To prevent it, we must ignore the field
point by excluding a small sphere, Σ, of radius  centered around P .
This exclusion, introduces an additional term on either sides of the
above equation. Further, let us call the remaining surface S. Thus,
Z   Z  
1 1
φ1 ∇ · ndA + φ1 ∇ · ndA =
Σ |x − x0 | S |x − x0 |
Z   Z  
1 1
0|
∇ φ1 · ndA + ∇ φ1 · ndA
Σ |x − x S |x − x0 |

We can approximate φ1 and its gradient to be constant over Σ so that

the first terms on the either sides of the equation are approximated

80
as
 
−1 2
Z Z
1
φ1 ∇ · ndA = φ1 (x) sin θdθdϕ
Σ |x − x0 | 2
= −4πφ1 (x)
Z   Z
1 1 2
0|
∇ φ 1 · ndA = ∇ φ 1 (x) ·  sin θdθdϕ
Σ |x − x 

In the limit  → 0, the second of the above two terms goes to zero so
that,
Z   Z  
1 1
−4πφ1 (x) + φ1 ∇ · ndA = ∇ φ1 · ndA
S |x − x0 | S |x − x0 |
on rearrangement,
Z   Z  
1 1 1 1
φ1 (x) = φ1 ∇ · ndA − ∇ φ1 · ndA
4π S |x − x0 | 4π S |x − x0 |

If φ1 (x) = κ on the surface S, then,

Z  
1 1
φ1 (x) = κn · ∇ dA
4π S |x − x0 |

If µ = κn, Z  
1 1
φ1 (x) = µ · n∇ dA
4π S |x − x0 |
Thus, any solution of Laplace’s equation, can be expressed as a po-
tential due to source doublets on a surface. A solution of Laplace’s
equation is essentially a potential of an irrotational, solenoidal veloc-
ity.

81
Chapter 3

Equations governing the

motion of a fluid

3.1 Material integrals in a moving fluid

• We prove that
dτ ∗
= (∇· u)τ ∗ , (3.1.1)
dt
where
δτ (t)
τ∗ = lim (3.1.2)
δτ (t0 )
δτ (t0 )→0

We begin with the definition of the first derivative,

dτ ∗ τ (t + h) − τ (t)
= lim
dt h→0 h
Using (3.1.2),
dτ ∗
 
1 δτ (t + h) δτ (t)
= lim lim − lim
dt h→0 h δτ (t0 )→0 δτ (t0 ) δτ (t0 )→0 δτ (t0 )

or,
dτ ∗ 1 δτ (t + h) − δτ (t)
= lim lim ,
dt h→0 h δτ (t0 )→0 δτ (t0 )
But
dδτ
δτ (t + h) − δτ (t) = h ,
dt
which from equation (3.1.1) of the book is
δτ (t + h) − δτ (t) = h ∇· uδτ (t)
so that
dτ ∗ 1 h ∇· uδτ (t) ∇· uδτ (t)
= lim lim = lim lim
dt h→0 h δτ (t0 )→0 δτ (t0 ) h→0 δτ (t0 )→0 δτ (t0 )

Since the function no longer depends on h, the limit as h → 0 is ineffective

and
dτ ∗ δτ (t)
= ∇· u lim = (∇· u)τ ∗
dt δτ (t0 )→0 δτ (t0 )

82
• We will now argue that, if δl is a material line element, then
dδl
= δl · ∇ u + o(|δl|) (3.1.3)
dt
A material line element’s length changes only if the velocity of the fluid
at its either ends is not the same. Let u0 and u1 be the velocities of the
two ends. Then, for a small element, u1 = u0 + δl · ∇ u + o(|δl|) so that
the difference in velocities is δl · ∇ u + o(|δl|).

• The volume of a cylindrical volume element with end faces of area δS and
generator δl is δτ = δS · δl = δSi δli . Since we have
1 dδτ
lim = ∇· u,
δτ →0 δτ dt
we can as well write
dδτ
= ∇· uδτ + o(δτ ) (3.1.4)
dt
or
dδli dδSi ∂uj
δSi + δli = δSi δli + o(δτ )
dt dt ∂xj
Using (3.1.3) in Cartesian tensor form,

∂ui dδSi ∂uj

δlj δSi + δli = δSi δli + o(δτ )
∂xj dt ∂xj

Interchanging i and j in the first term,

∂uj dδSi ∂uj
δli δSj + δli = δSi δli + o(δτ )
∂xi dt ∂xj
or  
∂uj dδSi ∂uj
δli δSj + − δSi = o(δτ )
∂xi dt ∂xj
If this were to be true in general then,
dδSi ∂uj ∂uj
= δSi − δSj + o(|δS|) (3.1.5)
dt ∂xj ∂xi

We can use the above equation to obtain an expression for,

d dδSi dρ
(ρδSi ) = ρ + δSi
dt dt dt 
∂uj ∂uj
= ρ δSi − δSj + o(|δS|) +
∂xj ∂xi
 
∂uj
δSi −ρ
∂xj
∂uj
= −ρδSj + o(|δS|) (3.1.6)
∂xi

83
• From (3.1.3),
dδl
2δl · = 2δl · ∇ u · δl + o(|δl|2 )
dt
or,
dδl ∂uj
2δl = 2δlmi δlmj + o(|δl|2 )
dt ∂xi
or,
1 dδl ∂uj
= mi mj + o(|δl|), (3.1.7)
δl dt ∂xi
where we wrote δl = δlm, m being a unit vector along the material line
element. Similarly, from (3.1.6),
d ∂uj
δSi (ρδSi ) = −ρδSj δSi + o(|δS|2 )
dt ∂xi
d ∂uj
δS (ρδS) = −ρ(δS)2 nj ni + o(|δS|2 )
dt ∂xi
1 d ∂uj
(ρδS) = −ni nj + o(|δS|), (3.1.8)
ρδS dt ∂xi
where we wrote δS = δSn, n being a unit vector normal to the material
area element.
• Let us consider the change of a line integral over a material element. Let
Z Q
I= θdl,
P

where θ is an intensive property of the fluid, dependent only on x and t.

We first express the integral as a Riemann sum,
Z Q X
θdl = lim θn δln ,
P →0
n

where  is the size of the largest sub-interval. Thus,

d Q
Z
d X
θdl = lim θn δln
dt P dt →0 n
X D
= lim (θn δln )
→0
n
Dt
X  Dθn Dδln

= lim δln + θn
→0
n
Dt Dt

For a material element, the material derivative is same as total derivative

so that
d Q X  Dθn
Z 
dδln
θdl = lim δln + θn
dt P →0
n
Dt dt
Using (3.1.3),

d Q X  Dθn
Z 
θdl = lim δln + θn δl · ∇ u + θn o(|δl|)
dt P →0
n
Dt

84
 or, since o(|δl|) < 
Z Q X  Dθn 
d
θdl = lim δln + θn δl · ∇ u
dt P →0
n
Dt

or Z Q Z Q Z Q
d Dθ
θdl = dln + θdl · ∇ u (3.1.9)
dt P P Dt P

• Now consider the surface integral over a material element,

Z Z
I = θndS = θdSi

Writing the integral as a Riemann sum

(n)
X
I = lim θn δSi ,
→0
n

(n)
where  is the size of the largest area interval and δSi is the nth area
element. Thus,
dI d X (n)
= lim θn δSi
dt dt →0 n
X D  (n)

= lim θn δSi
→0
n
Dt
!
(n)
X Dθn (n) DδSi
= lim δSi + θn
→0
n
Dt Dt

For a material element, the material derivative is same as total derivative

so that !
(n)
dI X Dθn (n) dδSi
= lim δSi + θn
dt →0
n
Dt dt

Using (3.1.5),

dI X  Dθn (n) 
∂uj (n) ∂uj (n)

= lim δSi + θn δSi − δSj + o(|δS|)
dt →0
n
Dt ∂xj ∂xi

Since o(|δS|) < ,

dI X  Dθn (n) 
∂uj (n) ∂uj (n)

= lim δSi + θn δS − δS
dt →0
n
Dt ∂xj i ∂xi j

or Z Z Z Z
d Dθ ∂uj ∂uj
θdSi = dSi + θ dSi − θ dSj (3.1.10)
dt Dt ∂xj ∂xi

• An integral over a material volume element is

Z
I = θdτ,

85
 Expressed as a Riemann sum,
X
I = lim θn δτn ,
→0
n

where  is the size of the largest volume interval so that

dI d X
= lim θn δτn
dt dt →0 n
X D
= lim (θn δτn )
→0
n
Dt
X  Dθn Dδτn

= lim δτn + θn
→0
n
Dt Dt

For a material element, the material derivative is same as total derivative

so that
dI X  Dθn dδτn

= lim δτn + θn
dt →0
n
Dt dt
Using (3.1.1),
 
dI X Dθn
= lim δτn + θn ∇· uδτn
dt →0
n
Dt
or Z Z Z
d Dθ
θdτ = dτ + θ ∇· udτ (3.1.11)
dt Dt
• A useful form of (3.1.11) is
Z Z  
d D(θρ)
θρdτ = + θρ ∇· u dτ
dt Dt
Z  
Dθ Dρ
= ρ+θ + θρ ∇· u dτ
Dt Dt

From the equation of mass conservation,

Dρ
= −ρ ∇· u
Dt
so that Z Z
d Dθ
θρdτ = ρ dτ (3.1.12)
dt Dt
• Several conservation laws can be put in the form
Z Z
d
θρdτ = Qdτ
dt
Using (3.1.12), the left hand side becomes,
Z Z
Dθ
ρ dτ = Qdτ,
Dt
which can be true only if
Dθ
ρ =Q (3.1.13)
Dt

86
 • While deriving (3.1.13) we travelled with the material element. We can
get to the same conclusion if we, instead, focus on a fixed volume in space.
If V is such a volume with surface area A then
Z
d
ρθdV
dt
is the rate of change of the quantity ρθ contained in V . The change can
be either because of a flux through the boundary or a source (or a sink)
in V . If n is the outward normal,
Z Z Z
d
ρθdV = − θρu · ndA + QdV
dt
Since the volume of integration is fixed, we can interchange the order of
differentiation and integration on the left hand side to get
Z Z Z
∂
(θρ)dV = − θρu · ndA + QdV
∂t
Using the divergence theorem on the first term on the right,
Z Z Z
∂
(θρ)dV = − ∇·(θρu)dV + QdV (3.1.14)
∂t
which immediately leads to the differential form,
∂
(θρ)dV = − ∇·(θρu) + Q (3.1.15)
∂t

3.2 Equations of motion

• We note that the relation between time derivative of a vector A in a fixed
and a rotating frame of reference is
   
dA dA
= + Ω ∧ A, (3.2.1)
dt f dt r

where the subscripts f and r mean ‘fixed’ and ‘rotating’. Ω is the angular
velocity of the rotating frame of reference with respect to the fixed frame
of reference. An immediate consequence of this equation is
   
dΩ dΩ
= (3.2.2)
dt f dt r

• If A = x,    
dx dx
= + Ω ∧ x,
dt f dt r

Differentiating once more

 2         
d x d dx dΩ dx
2
= + ∧x + Ω∧
dt f dt dt r f dt f dt f

87
 or
d2 x
          
dx dx dΩ dx
= + Ω∧ + ∧x + Ω∧ + Ω∧x
dt f dt2 r dt r dt f dt r

or
d2 x d2 x
       
dΩ dx
= + ∧x + 2Ω ∧ + Ω ∧(Ω ∧ x) (3.2.3)
dt2 f dt2 r dt r dt r

where we used (3.2.2).

• The momentum density of a fluid element is ρu. The rate of change of
this quantity is Z
d
ρudτ,
dt
which using (3.1.12) can be written as
Z
Du
ρ dτ
Dt
If F is the body force per unit volume and σij is the surface force per unit
area. then the total force is
Z Z Z Z
∂σij
ρFi dτ + σij nj dS = ρFi dτ + dτ
∂xj
By Newton’s second law, this quantity is equal to the total force acting
on the fluid element, so that
Z Z Z
Dui ∂σij
ρ dτ = ρFi dτ + dτ, (3.2.4)
Dt ∂xj
which, in differential form is,
Dui ∂σij
ρ = ρFi + , (3.2.5)
Dt ∂xj

• We can get an equation of motion even by considering a fixed volume V

of the fluid. The rate of change of momentum of fluid in V is
Z
d
ρui dV,
dt
Since the volume of integration is fixed, we can as well write it as
Z
∂
(ρui )dV
∂t
The flux of momentum through the surface A of the volume is
Z
− ρui uj nj dA

while the ‘source terms’ are

Z Z
ρFi dV + σij nj dA

88
 so that the equation of motion is
Z Z Z Z
∂
(ρui )dV = − ρui uj nj dA + ρFi dV + σij nj dA (3.2.6)
∂t
Using divergence theorem for the last term on the right hand side,
Z Z Z Z
∂ ∂σij
(ρui )dV = − ρui uj nj dA + ρFi dV + dV (3.2.7)
∂t ∂xj

• If F is a conservative force so that ρF = − ∇(ρΨ), for a potential function

Ψ, the second term on the right hand side of (3.2.6) can be written as
Z Z Z
∂ρΨ
ρFi dV = − dV = − ρΨni dA
∂xi
so that
Z Z Z
∂
(ρui )dV = − ρui uj nj dA + (−ρΨni + σij nj ) dA
∂t
If the motion is steady, the integrand on the left hand side is zero and
Z Z
ρui uj nj dA = (−ρΨni + σij nj ) dA (3.2.8)

Equation (3.2.8) says that under steady state the convective flux of mo-
mentum out of the surface A bounding V is balanced by the resultant
contact force exerted at the boundary by the surrounding medium and
the resultnt force at the boundary arising from the stress system equiva-
lent to the body force. The loss of momemtum is exactly compensated by
the two forces.

3.3 The expression for the stress tensor

• The average value of the normal component of stress on a surface element
at x, over all directions of n is
Z
1
hσi = ni σij nj dΩ(n)
4π
Since we are changing only the directions of the normals, staying at the
same point x, we can pull out σij out of the integral,
Z
1
hσi = σij ni nj dΩ(n)
4π
Using (4.10.1),
1 4π 1
hσi = σij δij = σii (3.3.1)
4π 3 3
• For any tensor {σij }, the quantity σii is an invariant under rotations.
Therefore, from (3.3.1), hσi is an invariant quantity. If the fluid were
static, σij = −pδij and hence hσi = −p. Thus, the analog of pressure, in
general flow conditions is −hσi.

89
• Taking the analogy in the reverse direction, we define the pressure at a
point in a moving fluid as
σii
p=− (3.3.2)
3
• The pressure so defined is not the same as the thermodynamic pressure.
The latter quantity is defined for equilibrium conditions while the former,
being a purely mechanical quantity, is defined for fluids in motion. It is
only when the fluid is at rest does the mechanical pressure defined by
(3.3.2) become identical with thermodynamic pressure.
• We write the stress tensor as a sum of an isotropic tensor and a ‘deviatoric’
tensor. Thus,
σij = −pδij + dij
The deviatoric part, dij is assumed to be proportional to the local velocity
gradient. The most general linear relationship between the two tensors dij
and ∇ u is
∂uk
dij = Aijkl
∂xl
Since we wrote
   
∂uk 1 ∂uk ∂ul 1 ∂uk ∂ul
= + + − = ekl + ξkl
∂xl 2 ∂xl ∂xk 2 ∂xl ∂xk
and we further noticed that ξkl is an anti-symmetric tensor so that it can
be expressed in terms of a single vector ω as
1
ξkl = − klm ωm
2
and hence
1
dij = Aijkl ekl − Aijkl klm ωm
2
The tensor Aijkl is a characteristic of the fluid alone. If the fluid is isotropic
then so is Aijkl . Now, an isotropic tensor of fourth order can be written,
following (4.12.18), as
Aijkl = µδik δjl + µ0 δil δjk + µ00 δij δkl ,
where µ, µ0 and µ00 are scalars. Since the stress tensor is symmetric, µ = µ0
so that
Aijkl = µ(δik δjl + δil δjk ) + µ00 δij δkl ,
Therefore,
1
dij = Aijkl ekl − Aijkl klm ωm
2
= µ(δik δjl + δil δjk )ekl + µ00 δij δkl ekl −
µ µ00 δij δkl
(δik δjl + δil δjk ) klm ωm − klm ωm
2 2
µ µ00
= 2µeij + µ00 ekk δij − (ijm + jim )ωm − δij kkm ωm
2 2
Now, ijm + jim = 0 and kkm = 0, so that
dij = 2µeij + µ00 ∆δij (3.3.3)

90
 • Since the deviatoric tensor dij does not contribute to the normal stress,
dii = 0. From (2.3.3),

dii = 2µeii + µ00 ∆δii = 2µ∆ + 3µ00 ∆ = ∆(2µ + 3µ00 )

dii = 0 implies
2
µ00 = − µ,
3
so that (3.3.3) becomes,
 
1
dij = 2µ eij − ∆δij (3.3.4)
3

• Molecular relaxation times involving mass transport are usually of the or-
der of 10−9 s1 . Time scale corresponding to macroscopic velocity gradient
is reciprocal of | ∇ u|. Thus, only when | ∇ u| is of O(109 ) does molec-
ular motion affect the continuum variables. That is the reason why the
linear relationship between dij and ∇ u is valid for a very large range of
gradients.
• Using (3.3.4) in the expression for the stress tensor, we get
 
1
σij = −pδij + 2µ eij − ∆δij (3.3.5)
3

Putting it in the equation of motion (3.2.5),

  
Dui ∂p ∂ 1
ρ = ρFi − + 2µ eij − ∆δij , (3.3.6)
Dt ∂xi ∂xj 3

Since  
1 ∂ui uj
eij = + ,
2 ∂xj xi
the divergence of rate-of-strain tensor is
 2    
1 ∂ ui ∂ ∂uj 1 2 ∂∆
+ = ∇ ui +
2 ∂xj ∂xj ∂xi ∂xj 2 ∂xi

Therefore, the equation of motion is

 
Dui ∂p 1 ∂∆
ρ = ρFi − + µ ∇2 ui + (3.3.7)
Dt ∂xi 3 ∂xi

This is the Navier-Stokes equation. If the fluid is incompressible, it is

simplified to
Dui ∂p
ρ = ρFi − + µ∇2 ui
Dt ∂xi
and be expressed in Gibbs notation as
Du
ρ = ρF − ∇ p + µ∇2 u (3.3.8)
Dt
1 See p.20 of Molecular Relaxation in Liquids, by Biman Bagchi

91
 • Equation (1.3.2) of the book defined the vector Σ(n) as the force exerted
by the fluid on the side of the surface element to which n points, on the
fluid on the side which n points away from. Further, equation (1.3.5) of
the book defined the stress tensor as Σj = σij nj . σij is the i component
of the force per unit area exerted on a plane surface whose normal is in
the j direction. If we consider a simple shearing motion with velocity
(U (y), 0, 0), the only non-zero components of the deviatoric stress tensor,
in an incompressible fluid, are
dU
d12 = d21 = µ
dy
d12 is the x component of the force per unit area on a plane surface element
whose normal points in the positive y direction. Further, it is the force
exerted by the fluid on the side to which the normal points on the fluid
on the side which n points away from. Therefore, if we consider an area
element in the xz plane, then the fluid in the upper part (y > 0) exerts
a force in the positive x direction. Clearly, the force is in a direction to
erase the velocity difference between a layer of fluid in the xz plane and
the one just above it. This can happen only if µ > 0.

• Consider an interface separating two media. Referring to figure (1.9.4)

of the book, if ti denotes the tangent to the interface then the tangential
00
stress in upper medium is ti σij nj . Continuity of tangential stress implies
00 0
ti σij nj = ti σij nj (3.3.9)

Using (1.7.8) for the normal components

 
0 00 1 1
ni σij nj − ni σij nj = γ + (3.3.10)
R1 R2

3.3.1 Exercise
1. Continuing to refer to figure (1.9.4) of the book, if the upper medium is a
gas, then the continuity of tangential stress, given by (3.3.9) becomes
0
0 = ti σij nj

Thus, if there is a material line element normal to the interface, then there
is no tangential force acting on it. Therefore, it will continue to be in the
normal direction.
(Note that σij is the force per unit area in the i direction on a small
material area element, normal to which points in the j direction. Further,
it is the force by the fluid in which the normal points, on the fluid on
the side which the normal points away from. If we choose a coordinate
system such that the interface lies in the xy plane and the z axis points
upwards then σ12 n2 is the force in the x direction and σ12 n1 is the force
in the y direction. To get their magnitudes, we just need to take a dot
product with a unit vector in that direction. Thus, the magnitude of the
two forces in xy plane are n1 σ12 n2 and n2 σ12 n1 .)

92
3.4 Changes in the internal energy of a fluid in
motion
• A material element of a fluid in motion is not a thermodynamic system
in equilibrium. Therefore, it is not right to assign thermodynamic vari-
ables to it without close examination. Of the thermodynamical variables,
density ρ can be defined as a local mass per unit volume even if the ma-
terial element is not in equilibrium. The first law of thermodynamics is
applicable even to systems not in equilibrium. Since the heat added to the
element dQ¯ and the work done on it dW
¯ are experimental quantities those
can be observed, so is their sum dU . Therefore, even internal energy can
be defined for a material element.
• If the fluid is homogeneous, then we can use the instantenous values of
ρ and U to define other thermodynamic quantities. For instance, we can
define T using the equation of state.
• Consider a fluid element of volume τ and surface area S. The volume
forces Fi do a work on it at the rate
Z
ui Fi dτ

while the surface forces σij do work at the rate

Z Z
∂ui σij
ui σij nj dS = dτ
∂xj

Thus, the rate at which all forces do work on the fluid is

Z Z
∂ui σij
= ui Fi dτ + dτ
∂xj
Z Z Z
∂σij ∂ui
= ui Fi dτ + ui dτ + σij dτ
∂xj ∂xj

The second term above arises because of a variation in stress across the
fluid element.It leads to change in the element’s kinetic energy. The third
term above arises because of a variation in velocity across the fluid element.
It causes the element to deform without a change in the element’s kinetic
energy. The work done in deformation shows as an increase in internal
energy of the element. The three integrals can be written as one with the
integrand,
 
∂σij ∂ui ∂σij ∂ui
ui Fi + ui + σij = ui Fi + + σij
∂xj ∂xj ∂xj ∂xj

Using (3.2.5) the terms in the bracket can be combined to get

Dui ∂ui
ρui + σij (3.4.1)
Dt ∂xj

The integrand, written above, can be interpreted as a the rate at which

work is being done on a fluid element per unit volume. To get the rate

93
 per unit mass, we divide it be ρ. Thus, the rate at which work is done by
volume and surface forces per unit mass is
Dui σij ∂ui 1 D(ui ui ) σij ∂ui
ui + = + (3.4.2)
Dt ρ ∂xj 2 Dt ρ ∂xj
The form on the right hand side makes it clear that the first term repre-
sents the change in kinetic energy while the second term represents the
change in internal energy.
• The rate at which heat enters a fluid element due to temperature gradient
is Z Z
k ∇ T · ndS = ∇·(k ∇ T )dτ

Thus the rate, per unit mass, of addition of heat to the element is
 
1 ∂ ∂T
k (3.4.3)
ρ ∂xi ∂xi

• If u is the internal energy per unit mass, then by first law of thermody-
namics, the rate of its change per unit mass is
 
Du σij ∂ui 1 ∂ ∂T
= + k (3.4.4)
Dt ρ ∂xj ρ ∂xi ∂xi
or    
Du 1 σij ∂ui σij ∂ui 1 ∂ ∂T
= + + k
Dt 2 ρ ∂xj ρ ∂xj ρ ∂xi ∂xi
We can always interchange the indices in the second term in the bracket,
it is a scalar after all, to get
   
Du 1 σij ∂ui σji ∂uj 1 ∂ ∂T
= + + k
Dt 2 ρ ∂xj ρ ∂xi ρ ∂xi ∂xi
But the stress tensor is symmetric, so that
   
Du 1 σij ∂ui σij ∂uj 1 ∂ ∂T
= + + k
Dt 2 ρ ∂xj ρ ∂xi ρ ∂xi ∂xi
or  
Du σij 1 ∂ ∂T
= eij + k (3.4.5)
Dt ρ ρ ∂xi ∂xi
Using (3.3.5),
    
Du eij 1 1 ∂ ∂T
= −pδij + 2µ eij − ∆δij + k
Dt ρ 3 ρ ∂xi ∂xi
or     
Du eii µ 1 1 ∂ ∂T
= −p +2 eij eij − ∆eii + k
Dt ρ ρ 3 ρ ∂xi ∂xi
Putting eii = ∆,
   
Du ∆ µ 1 2 1 ∂ ∂T
= −p + 2 eij eij − ∆ + k (3.4.6)
Dt ρ ρ 3 ρ ∂xi ∂xi

94
• We will now show that
∆2
      
∆δij ∆δij ∆δij
−p∆+2µ eij eij − = −pδij +2µ eij − eij −
3 3 3 3

    
∆δij ∆δij ∆δij
RHS = −pδij + 2µ eij − eij −
3 3 3
∆2 δij δij
 
δij δij ∆δij
= −p∆ + 2µ eij eij − 2eij +
3 3 9
2
 
2 ∆
= −p∆ + 2µ eij eij − ∆eii +
3 3
 
1
= −p∆ + 2µ eij eij − ∆2
3
= LHS

• In (3.4.6), the first term on the right hand side involves only the isotropic
part of stress and rate-of-strain while the second term has only the cor-
responding deviatoric parts. Further, the second term is non-negative.
It represents the heating of the element due to frictional forces. It is a
quantity of interest and therefore deserves a separate symbol
 
µ 1 2
Φ=2 eij eij − ∆ (3.4.7)
ρ 3

• We have so far argued that the usual definitions of ρ and U can be used
in non-equilibrium states of the fluid element. We denote the pressure
obtained from using instantaneous values of ρ and U in equilibrium equa-
tions of state by pe . Recall that p, as used in the equations above, is
defined to be the average stress at a point. In general p and pe are not the
same when the fluid is in motion, although they are identical in a static
fluid.
• Since the difference between p and pe is solely due to fluid’s motion, we
assume that p−pe depends only on the local velocity gradient. We further
assume that the dependence is linear so that
∂ui 1
p − pe = Bij = Bij eij − Bij ijk ωk
∂xj 2

If the fluid is isotropic, we can expect Bij to be an isotropic tensor. From

section 4.12, an isotropic sencond rank tensor is a scalar multiple of Kro-
necker delta. Therefore, let

Bij = −κδij (3.4.8)

so that
p − pe = −κ∆ (3.4.9)
The constant κ is usually called the bulk viscosity of the fluid.

95
• Since p and pe are identical in a static fluid, we can surmise that p − pe
is due to deviatoric part of the stress. If the fluid contracts during the
motion, that is its local ∆ is negative, p will be greater than pe . Therefore,
we can write p − pe = −κ∆, assuming that κ > 0. If the fluid expands
during the motion, then its local ∆ is positive and p will end up being
smaller than pe . Therefore, once again p − pe = −κ∆ is a valid relation if
κ > 0.
• We can now write the first term on the right hand side of (3.4.6) as

∆ ∆ ∆2
−p = −pe + κ (3.4.10)
ρ ρ ρ
The term
∆
−pe
ρ
is contribution by the reversible effects of equilibrium pressure while the
term
∆2
κ ≥0
ρ
is the dissipative effect of bulk viscosity. Although the second term is
usually much smaller than the first one on the right hand side of (3.4.10),
being a positive quantity, periodic pressure variations over a long enough
time can make it significant. That is why, it is important in transmission
of sound waves in fluids.
• We will now derive equation (3.4.11) of the book. Starting from (3.4.6)
and (3.4.7),  
Du ∆ 1 ∂ ∂T
= −p + Φ + k
Dt ρ ρ ∂xi ∂xi
From (3.4.10),

∆2
 
Du ∆ 1 ∂ ∂T
= −pe + κ +Φ+ k
Dt ρ ρ ρ ∂xi ∂xi

We can write the mass conservation equation as

1 Dρ
∆=−
ρ Dt
so that
∆2
 
Du pe Dρ 1 ∂ ∂T
= 2 +κ +Φ+ k
Dt ρ Dt ρ ρ ∂xi ∂xi
or
∆2
   
Du D 1 1 ∂ ∂T
= −pe +κ +Φ+ k
Dt Dt ρ ρ ρ ∂xi ∂xi
or, since volume per unit mass v = 1/ρ,

∆2
 
Du Dv 1 ∂ ∂T
= −pe +κ +Φ+ k
Dt Dt ρ ρ ∂xi ∂xi

96
 or
∆2
 
Du Dv 1 ∂ ∂T
+ pe =κ +Φ+ k
Dt Dt ρ ρ ∂xi ∂xi
or
∆2
 
Ds 1 ∂ ∂T
T =κ +Φ+ k (3.4.11)
Dt ρ ρ ∂xi ∂xi
where we have used the first part of equation (1.5.20) of the book, namely,
T ds = du + pe dv. If we also use the second part, that is T ds = cp dT −
βvT dpe , we get
∆2
 
DT Dpe 1 ∂ ∂T
cp − βvT =κ +Φ+ k
Dt Dt ρ ρ ∂xi ∂xi
If we were to use ρ throughout,
∆2
 
DT βT Dpe 1 ∂ ∂T
cp − =κ +Φ+ k (3.4.12)
Dt ρ Dt ρ ρ ∂xi ∂xi

3.5 Bernoulli’s theorem

• We will derive equation (3.5.1) of the book. To do so, we begin with the
equation of motion (3.2.5)
Dui ∂σij
ρ = ρFi +
Dt ∂xj
which is same as
Dui ui ∂σij
ui = ui Fi +
Dt ρ ∂xj
Now,
∂ ∂σij ∂ui
(ui σij ) = ui + σij
∂xj ∂xj ∂xj
so that
Dui 1 ∂(ui σij ) σij ∂ui
ui = ui Fi + −
Dt ρ ∂xj ρ ∂xj
or
D  ui ui  1 ∂(ui σij ) σij ∂ui
= ui Fi + − (3.5.1)
Dt 2 ρ ∂xj ρ ∂xj
From (3.4.4)  
Du σij ∂ui 1 ∂ ∂T
= + k
Dt ρ ∂xj ρ ∂xi ∂xi
so that  
σij ∂ui Du 1 ∂ ∂T
− =− + k
ρ ∂xj Dt ρ ∂xi ∂xi
Putting this in (3.5.1), we get
 
D  ui ui  1 ∂(ui σij ) Du 1 ∂ ∂T
= ui Fi + − + k
Dt 2 ρ ∂xj Dt ρ ∂xi ∂xi
or  
D  ui ui  1 ∂(ui σij ) 1 ∂ ∂T
u+ = ui Fi + + k (3.5.2)
Dt 2 ρ ∂xj ρ ∂xi ∂xi

97
• If F is a conservative force with a potential function Ψ,
∂Ψ DΨ ∂Ψ
ui Fi = −ui =− +
∂xi Dt ∂t
If Ψ is also independent of time,
DΨ
ui Fi = −
Dt
Putting it in (3.5.2), we get
 
D  ui ui  1 ∂(u σ ) 1 ∂
i ij ∂T
+u+Ψ = + k (3.5.3)
Dt 2 ρ ∂xj ρ ∂xi ∂xi

From (3.3.5)  
1
σij = −pδij + 2µ eij − ∆δij
3
so that  
∆
ui σij = −puj + 2µ ui eij − uj
3
and hence,
 
∂(ui σij ) ∂(puj ) ∂ ∆
= − + 2µ ui eij − uj
∂xj ∂xj ∂xj 3
 
∂uj ∂p ∂ ∆
= −p − uj + 2µ ui eij − uj
∂xj ∂xj ∂xj 3
 
∂uj Dp ∂p ∂ ∆
= −p − + + 2µ ui eij − uj
∂xj Dt ∂t ∂xj 3

and hence, from (3.5.3),

  
D  ui ui  1 ∂uj Dp ∂p ∂ ∆
+u+Ψ = −p − + + 2µ ui eij − uj
Dt 2 ρ ∂xj Dt ∂t ∂xj 3
 
1 ∂ ∂T
+ k
ρ ∂xi ∂xi

Using mass conservation equation,

∂uj 1 Dρ
=−
∂xj ρ Dt

so that
  
D u ui i
 p Dρ 1 Dp 1 ∂p ∂ ∆
+u+Ψ = − + + 2µ ui eij − uj
Dt 2 ρ2 Dt ρ Dt ρ ∂t ∂xj 3
 
1 ∂ ∂T
+ k
ρ ∂xi ∂xi

Now,    
p Dρ 1 Dp D 1 1 Dp D p
− = −p − =− ,
ρ2 Dt ρ Dt Dt ρ ρ Dt Dt ρ

98
 so that
    
D u ui i
 D p 1 ∂p ∂ ∆
+u+Ψ = − + + 2µ ui eij − uj
Dt 2 Dt ρ ρ ∂t ∂xj 3
 
1 ∂ ∂T
+ k
ρ ∂xi ∂xi
or
    
D ui ui p 1 ∂p ∂ ∆
+u+ +Ψ = + 2µ ui eij − uj
Dt 2 ρ ρ ∂t ∂xj 3
 
1 ∂ ∂T
+ k
ρ ∂xi ∂xi

If the pressure is steady,

     
D ui ui p 2µ ∂ ∆ 1 ∂ ∂T
+u+ +Ψ = ui eij − uj + k
Dt 2 ρ ρ ∂xj 3 ρ ∂xi ∂xi
(3.5.4)
If the fluid is frictioness, µ = 0 and it is non-conducting, k = 0 so that
DH
= 0, (3.5.5)
Dt
where the function H is defined as
ui ui p
H= +u+ +Ψ (3.5.6)
2 ρ
If we denote the speed of a fluid parcel by q,

q2 p
H= +u+ +Ψ (3.5.7)
2 ρ

• Since T ds = du + pdv, T ∇ S = ∇ u + p ∇ v or
   
1 p 1
T ∇S = ∇u + p∇ = ∇u + ∇ − ∇ p,
ρ ρ ρ
or  
1 p
T ∇S + ∇p = ∇ u + (3.5.8)
ρ ρ
From (3.5.7)
q2
   
p
∇H = ∇ +Ψ +∇ u+
2 ρ
From (3.5.8), we get

q2
 
1
∇H = T ∇S + ∇ + Ψ + ∇p (3.5.9)
2 ρ

• For a steady flow of a frictionless fluid, the equation of motion (3.3.8)

becomes
ρu · ∇ u = −ρ ∇ Ψ − ∇ p

99
 Since
∇(u · u) = ∇ q 2 = 2u · ∇ u + 2u ∧ ∇ ∧ u,
we have
q2
 
∇ = u · ∇u + u∧ω
2
or
q2
 
ρ∇ − ρu ∧ ω = −ρ ∇ Ψ − ∇ p
2
or
q2
 
1
∇ + ψ + ∇p = u∧ω
2 ρ
Putting this relation in (3.5.9), we get

∇H = T ∇S + u∧ω (3.5.10)

• From the relation, T ds = du + pdv, we get

   
1 p 1
T ds = du + pd = du + d − dp
ρ ρ ρ

For an isentropic (adiabatic) process,

 
1 p
dp = d u +
ρ ρ

If  
2 ∂p
c = ,
∂ρ S
we have
dp = c2 dρ
or
1 c2
dp = dρ
ρ ρ
so that
c2
 
p
dρ = d u +
ρ ρ
or
c2
Z
1
u+ = dρ (3.5.11)
ρ ρ
or
c2
  Z
D 1 D
u+ = dρ (3.5.12)
Dt ρ Dt ρ
Using (3.5.11) in the definition (3.5.7) we get

q2
Z 2
c
H= + dρ + Ψ (3.5.13)
2 ρ

100
 • We will show that
1
− Ω ∧(Ω ∧ x) = ∇(Ω ∧ x)2 (3.5.14)
2
Starting from the right hand side,
1
∇(Ω ∧ x)2 = (Ω ∧ x) · ∇(Ω ∧ x) + (Ω ∧ x) ∧ ∇ ∧(Ω ∧ x) (3.5.15)
2
Since Ω is a constant,

∇ ∧(Ω ∧ x) = Ω ∇· x − Ω · ∇ x = 3Ω − Ω = 2Ω

so that

(Ω ∧ x) ∧ ∇ ∧(Ω ∧ x) = 2(Ω ∧ x) ∧ Ω = −2Ω ∧(Ω ∧ x) (3.5.16)

Now consider
 
∂
(Ω ∧ x) · ∇(Ω ∧ x) = ijk ei Ωj xk · el (pqr ep Ωq xr )
∂xl
∂
= ijk Ωj xk (pqr ep Ωq xr )
∂xi
= ijk Ωj xk pqr ep Ωq δir
= ijk Ωj xk pqi ep Ωq
= ijk pqi Ωj Ωq xk ep
= ijk ipq Ωj Ωq xk ep
= (δjp δkq − δjq δkp )Ωj Ωq xk ep
= xk Ωk Ωj ej − Ωj Ωj xk ek
= (x · Ω)Ω − Ω2 x

Thus,
(Ω ∧ x) · ∇(Ω ∧ x) = Ω ∧(Ω ∧ x) (3.5.17)
Putting (3.5.16) and (3.5.17) on the right hand side of (3.5.15),
1
∇(Ω ∧ x)2 = −Ω ∧(Ω ∧ x)
2

3.6 The complete set of governing equations

• Flow of a Newtonian fluid is described by
1. Mass conservation, (2.2.2),

1 Dρ
+ ∇· u = 0 (3.6.1)
ρ Dt

2. Momentum balance, (3.3.6)

  
Dui ∂p ∂ 1
ρ = ρFi − + 2µ eij − ∆δij (3.6.2)
Dt ∂xi ∂xj 3

101
 3. Energy balance, (3.4.11) and (3.4.12)
∆2
 
Ds DT βT Dpe 1 ∂ ∂T
T = cp − =κ +Φ+ k
Dt Dt ρ Dt ρ ρ ∂xi ∂xi
If we ignore effects of expansion damping, we can drop the first term
on the right hand side and replace pe with p to get
 
Ds DT βT Dp 1 ∂ ∂T
T = cp − =Φ+ k (3.6.3)
Dt Dt ρ Dt ρ ∂xi ∂xi
The six quantities ρ, u, p and T are the unknowns in these equations. Since
these are only five equations, we need one more, namely the equation of
state of the form
f (p, ρ, T ) = 0 (3.6.4)
to be able to solve for all the unknowns. The material parameters µ and
k are given functions of ρ and T .

3.6.1 Isentropic flows

• Recall that an isentropic flow is the one for which entropy of a fluid element
does not change throughout the course of its motion while a homentropic
flow is the one for which entropy per unit mass s is same throughout the
fluid.
• If we set the molecular transport coefficients, µ and k to zero, equation
(3.6.3) gives,
DT βT Dp
cp − (3.6.5)
Dt ρ Dt
Solving this equation gives T as a function of p. Coupled with the equation
of state, f (p, ρ, T ) = 0 gives ρ as a function of p. Now
Ds
=0
Dt
implies that s is a constant. To indicate that ρ is a function of p at
constant entropy, we write
ρ = ρ(p, s) (3.6.6)
If the flow were homentropic, we would have written ρ = ρ(p). For an
isentropic flow, at a fixed entropy of a fluid element, (3.6.6) gives
∂ρ ∂ρ ∂p
=
∂t ∂p ∂t
∂ρ
∇ρ = ∇p
∂p
therefore,
Dρ ∂ρ Dp
= (3.6.7)
Dt ∂p Dt
and hence equation (3.6.1) for mass conservation becomes
1 Dp
+ ∇· u = 0 (3.6.8)
ρc2 Dt

102
 where we have used the relation
 
2 ∂p
c = (3.6.9)
∂ρ s

Putting µ = 0 also simplifies the momentum balance equation (3.6.2) to

Du
ρ = ρF − ∇ p (3.6.10)
Dt

• Consider a fluid element at rest. Under the equilibrium condition, the

pressure gradient and the body force balance each other so that ρ0 F =
∇ p0 , where the subscript 0 indicates equilibrium values. Now let the fluid
be slightly perturbed so that the pressure becomes p = p0 + p1 and as a
result the density becomes ρ = ρ0 + ρ1 . Here, the subscript 1 indicates
the perturbed values. Now,
   
1 1 1 1 1 ρ1
= = = 1− ,
ρ ρ0 + ρ1 ρ0 1 + ρ1 /ρ0 ρ0 ρ0
up to first order in ρ1 . Further,
Dp Dp0 Dp1
= + ,
Dt Dt Dt
so that
    
1 Dp 1 ρ1 Dp0 Dp1 1 Dp0 Dp1 ρ1 Dp0
2
= 1− + = + − ,
ρc Dt ρ0 c2 ρ0 Dt Dt ρ0 c2 Dt Dt ρ0 Dt
up to first order terms in perturbed quantities. Under equilibrium condi-
tions,
Dp0 ∂p0
= = 0,
Dt ∂t
because the equilibrium velocity u0 = 0. Therefore,
1 Dp 1 Dp1 1 ∂p1
2
= 2
=
ρc Dt ρ0 c Dt ρ0 c2 ∂t
Similarly, ∇· u = ∇· u0 + ∇· u1 = ∇· u1 so that the equation of mass
conservation (3.6.8) becomes,
1 ∂p1
+ ∇· u1 = 0 (3.6.11)
ρ0 c2 ∂t
Now,
Du Du0 Du1
= +
Dt Dt Dt
Du1
=
Dt
∂u1
= + u · ∇ u1
∂t
∂u1
= ,
∂t

103
 up to first order terms in u1 . Therefore, the momentum balance equation
(3.6.10) becomes
∂u1
(ρ0 + ρ1 ) = (ρ0 + ρ1 )F − ∇ p0 − ∇ p1
∂t
or
∂u1
ρ0 = ρ1 F − ∇ p1 (3.6.12)
∂t
where we used the equilibrium condition ρ0 F = ∇ p0 . Differentiating
(3.6.11) with respect to t and substituting (3.6.12) in the result gives
1 ∂p1
= ∇2 p1 − ∇·(ρ1 F)
c2 ∂t
Now,
 
∂ρ1 F · ∇ p1
∇·(ρ1 F) = ρ1 ∇· F+F·∇ ρ1 = ρ1 ∇· F+F· ∇ p1 = ρ1 ∇· F+
∂p1 c2

so that
1 ∂p1 F · ∇ p1
2
= ∇2 p1 − ρ1 ∇· F − (3.6.13)
c ∂t c2
If the only external field is gravity, F = g so that ∇· g = 0 and hence,
1 ∂p1 g · ∇ p1
= ∇2 p1 −
c2 ∂t c2
For air under normal pressure, the velocity of sound c ≈ 330 ms−1 and
hence g/c2 ≈ 9 × 10−5 . Thus, the second term on the right hand side is
negligibly small and
1 ∂p1
= ∇ 2 p1 (3.6.14)
c2 ∂t
which is the equation of sound waves in fluids.

3.7 Incompressible flows

• Let the velocity u be such that it varies appreciably only over distances
comparable to L. In other words, variation in u over distances small
compared to L is negligible. Further, let the variation of u in space or
time be of a magnitude comparable to U . Then, the velocity field is
approximately solenoidal if
U
| ∇· u| 
L
or if  1 Dρ  U
 (3.7.1)
 
ρ Dt L


• If we choose ρ and s as independent variables, then we can express p =

p(ρ, s). In that case,
∂p ∂p
dp = dρ + ds
∂ρ ∂s

104
 and hence
Dp ∂p Dρ ∂p Ds
= +
Dt ∂ρ Dt ∂s Dt
or, writing in the usual thermodynamic convention,
 
Dp 2 Dρ ∂p Ds
=c + (3.7.2)
Dt Dt ∂s ρ Dt

Using this equation, we can write

 
Dρ 1 Dp 1 ∂p Ds
= 2 − 2
Dt c Dt c ∂s ρ Dt

and hence the condition for incompressibility (3.6.14) can be written as

 1 Dp  
1 ∂p Ds  U
− 2  (3.7.3)

 2
ρc Dt ρc ∂s ρ Dt L

This relation is valid if each of the two terms on the left hand side have
a magnitude small when compared with U/L. We will examine the two
terms separately.
• For the moment, assume that the flow is isentropic, in which case, the
equation of motion (3.6.10) can be written as
Du
∇ p = ρF − ρ
Dt
so that
q2
 
Du D
u · ∇ p = ρu · F − ρu · = ρu · F − ρ
Dt Dt 2
and
q2
 
Dp ∂p ∂p D
= + u · ∇p = + ρu · F − ρ
Dt ∂t ∂t Dt 2
due to which the first term on the left hand side of (3.7.5) becomes
1 D q2
 
1 Dp 1 ∂p 1
= + u · F −
ρc2 Dt ρc2 ∂t c2 c2 Dt 2
and its smallness means
 1 ∂p u · F 1 Dq 2  U
+ 2 − 2  (3.7.4)

 2
ρc ∂t c 2c Dt L
1. Smallness of  1 Dq 2 
 
 2
2c Dt

means that the ratio
U2 U
1⇒ 1
c2 c
The dimensionless quantity U/c is called Mach number. If the Mach
number of a flow is very small, then other terms on the left hand side
of (3.7.2) and (3.7.1) also being small, the fluid can be considered
effectively incompressible.

105
 2. Dimension of pressure is same as ρq 2 . If U is the typical variation in
velocity, L is the typical length scale and n the dominant frequency,
Ln too has dimensions of velocity and U Ln has dimensions of q 2 .
Thus the pressure change is of the order of ρU Ln and rate of change
of pressure is of the order of ρU Ln2 . Therefore,
 1 ∂p  U  1  U
 2  ⇒  2 ρU Ln2  
   
ρc ∂t L ρc L
which is same as
L2 n2
1 (3.7.5)
c2
This condition will be obviously violated if L is of the order of the
wavelength of a sound wave and n is of the order of its frequency.
Thus, compressibility cannot be ignored if the flow is caused due to
passage of sound.
3. Smallness of u · F U
 2 
 
c L
in the case of gravitational fields means
gL
1
c2
Now for air, c2 = γp/ρ (refer to Wikipedia) so that the smallness of
the second term on the left hand side of (3.7.2) is equivalent to
ρgL
 1,
γp
where γ is the adiabatic ratio. For air,
p
≈ 8 km
ρg
(refer to p. 20 of the book) so that the relation
ρgL
 1,
γp
is valid for if L is much lesser than γ = 11.2 kilometer, which means
always at the scale of the lab.
• We will now consider the term
 1  ∂p  Ds 
T2 =  2
 
ρc ∂s ρ Dt


From (3.6.19),  
1 ∂p βT
=
ρc2 ∂s ρ cp
so that  βT Ds   β Ds 
T2 =  = T
   
cp Dt cp Dt


106
From (3.6.3),
β  1 ∂

∂T
 
T2 =  Φ+ k
 
cp ρ ∂xi ∂xi


Now,
β β µ U2
Φ=
cp cp ρ L2
and  
β 1 ∂ ∂T k θ
k =β
cp ρ ∂xi ∂xi ρcp L2
Thus T2  U/L is possible is

βU µ
 1 (3.7.6)
cp ρL
βθκ
 1, (3.7.7)
LU
where
k
κ=
ρcp
is the thermometric conductivity.

107
Chapter 4

Some mathematical results

4.1 Green’s theorem

We will prove Green’s theorem
Z Z
2 2


φ∇ ψ − ψ∇ φ dV = (φ ∇ ψ − ψ ∇ φ) · ndA (4.1.1)

Let A = φ ∇ ψ and B = ψ ∇ ψ. Then,

Z Z
(A − B) · ndA = ∇· (A − B)dV

Now, ∇· A = ∇· (φ ∇ ψ) = ∇ φ·∇ ψ+φ∇2 ψ. Similarly, ∇· B = ∇ ψ·∇ φ+ψ∇2 φ.

Therefore, Z Z
φ∇2 ψ − ψ∇2 φ dV


(A − B) · ndA =

4.2 Solution of Poisson equation

We will show that the solution of the Poisson’s equation ∇2 φ = ∆ is
∆(x0 )
Z
1
φ(x) = − dV (x0 )
4π |x − x0 |

We begin with Fourier transform of ∇2 φ = ∆(x). Thus,

Z ∞ Z ∞
1 −ik·x 2 1
e ∇ φdV = e−ik·x ∆(x)dV
(2π)3/2 −∞ (2π)3/2 −∞

Use Green’s theorem (4.1.1) on the left hand side to get,

Z ∞ Z ∞
1 2 −ik·x 1
φ∇ e dV = e−ik·x ∆(x)dV
(2π)3/2 −∞ (2π)3/2 −∞
Z ∞
1
+ e−ik·x ∇ φ · ndA
(2π)3/2 −∞
Z ∞
1
− φ ∇ e−ik·x · ndA
(2π)3/2 −∞

108
The second and the third term on the left hand side can be written as
Z π Z 2π
R2
e−ik·x ∇ φ − φ ∇ e−ik·x sin θdθdϕ


lim 3/2
R→∞ (2π) 0 0

For a function φ that goes to zero faster than 1/r2 as r → ∞, the above
expression will be zero. Therefore,
Z ∞ Z ∞
1 2 −ik·x 1
φ∇ e dV = e−ik·x ∆(x)dV (4.2.1)
(2π)3/2 −∞ (2π)3/2 −∞

To calculate ∇2 e−ik·x , we assume, without loss of generality, that k is along the

positive z axis so that k · x = kr cos θ. Therefore,
 
2 −ik·x 1 ∂ 2 ∂ −ikr cos θ
∇ e = r e +
r2 ∂r ∂r
 
1 ∂ ∂ −ikr cos θ
sin θ e +
r2 sin θ ∂θ ∂θ
1 ∂ 2 −ikr cos θ
e
r2 sin2 θ ∂φ2
or
∇2 e−ik·x = −k 2 e−ikr cos θ = −k 2 e−ik·x
Therefore, equation (4.2.1) becomes,
Z ∞ Z ∞
k2 −ik·x 1
− φe dV = e−ik·x ∆(x)dV
(2π)3/2 −∞ (2π)3/2 −∞

If fˆ(k) denotes the Fourier transform of f (x), the above equation means,

ˆ
∆(k)
ˆ
−k 2 φ̂(k) = ∆(k) ⇒ φ̂(k) = −
k2
Therefore,
Z ∞ ˆ
1 ∆(k)
φ(x) = φh (x) − eik·x dVk
(2π)3/2 −∞ k2
where dVk denotes the volume element in the k-space and φh (x) is any solution
ˆ
of the homogeneous equation ∇2 φ = 0. Using the expression for ∆(k), we get
∞ ∞ 0
eik·(x−x ) 0
Z Z
1 0
φ(x) = φh (x) − ∆(x ) dV dVk ,
(2π)3 −∞ −∞ k2

where dV 0 is integration over x0 . Let,

∞ 0
eik·(x−x )
Z
1
G(x, x0 ) = − dVk
(2π)3 −∞ k2

be the Green function so that,

Z ∞
φ(x) = φh (x) + G(x, x0 )∆(x0 )dV 0 (4.2.2)
−∞

109
The Green function is an integral in k-space. Therefore, without loss of gener-
ality, we can orient the kz axis parallel to the vector x − x0 so that
∞ π 2π 0
eik|x−x | cos θ 2
Z Z Z
0 1
G(x, x ) = − k sin θdkdθdϕ
(2π)3 0 0 0 k2
or Z ∞ Z π
0 1 0
G(x, x ) = − eik|x−x | cos θ sin θdkdθ
(2π)2 0 0

Put t = cos θ so that,

Z ∞ Z −1
0 1 0
G(x, x ) = eik|x−x |t dtdk
(2π)2 0 1

or !
∞ 0
eik|x−x |t −1
Z
0 1
G(x, x ) = dk
(2π)2 ik|x − x0 | 1

0

or 0 0
∞
eik|x−x | − e−ik|x−x |
Z
i
G(x, x0 ) = dk
(2π)2 0 k|x − x0 |
or,
∞ 0 0
eik|x−x | − e−ik|x−x |
Z
i 1
G(x, x0 ) = dk
(2π)2 |x − x0 | 0 k
Now,
∞ 0 0
eik|x−x | − e−ik|x−x |
Z
dk =
0 k
Z ∞ ik|x−x0 | Z ∞ −ik|x−x0 |
e e
dk − dk =
0 k 0 k
Z ∞ ik|x−x0 | Z −∞ ik|x−x0 |
e e
dk − d(−k) =
0 k 0 (−k)
Z ∞ ik|x−x0 | Z 0 ik|x−x0 |
e e
dk + dk =
0 k −∞ k
Z ∞ ik|x−x0 |
e
dk
−∞ k

so that,
∞ 0
eik|x−x |
Z
0 i 1
G(x, x ) = dk (4.2.3)
(2π)2 |x − x0 | −∞ k
We can evaluate the integral using residue theorem. Consider a semi-circular
contour with diameter along the real axis, from −R to R and arc in the first
two quadrants. Now the integrand has a singularity the origin, a point on the
contour. Therefore, we have to take the Cauchy principal value of the integrand.
To that end, indent the contour by a small semi-circulararc with diameter along
the real axis, from −a to a and arc in the first two quadrants. The indented
contour now no longer has the singularity in its interior and hence by Cauchy

110
theorem, the integral around it is zero. If C1 denotes the larger arc and C2 the
smaller one, then
Z Z −a Z Z R ! ik|x−x0 |
e
+ + + dk = 0 (4.2.4)
C1 −R C2 a k

In the limit R → ∞, the integral along C1 vanishes because the integrand itself
goes to zero. Along C2 , k = ρeiθ , dk = iρeiθ dθ and the limits of the integral are
from π to 0. Thus,
0 0
eik|x−x | iρeiθ dθ
Z Z
dk = exp (iρ|x − x0 |(cos θ + i sin θ))
C2 k π ρeiθ
Z 0
= i exp (iρ|x − x0 |(cos θ + i sin θ)) dθ
π

In the limit ρ → 0, we can approximate the integrand by its Maclaurin series so

that,
0 0
eik|x−x |
Z Z
dk = i (1 + iρ|x − x0 |(cos θ + i sin θ)) dθ
C2 k π
= −iπ + 2ρ|x − x0 |,

which, in the limit of vanishing ρ gives −iπ. Therefore, equation (4.2.4) becomes,
∞ 0 0
eik|x−x | eik|x−x |
Z Z
dk = − dk = iπ
−∞ k C2 k

Therefore, equation (4.2.3) becomes,

1 1
G(x, x0 ) = −
4π |x − x0 |

Therefore from (4.2.2)

∞
∆(x0 )
Z
1
φ(x) = φh (x) − dV 0
4π −∞ |x − x0 |

Since φh (r) = 0 is a solution of the Laplace’s equation,

Z ∞
1 ∆(x0 )
φ(x) = − dV 0
4π −∞ |x − x0 |

is a solution of the Poisson’s equation ∇2 φ = ∆(x).

4.3 Another form of Stokes’s theorem

We will prove an analog of Stokes’ theorem for a scalar field,
I Z
f dl = − (∇ f ∧ n)dA (4.3.1)

111
Let F(x) = αf (x), where α is a constant vector. Therefore, Stokes’ theorem
for F is I Z
F · dl = ∇ ∧ F · ndA

that is I Z
α· f (x)dl = (∇ f (x) ∧ α) · ndA,
or, I Z
α· f (x)dl = − (α ∧ ∇ f (x)) · ndA,

Interchanging the dot and the cross products on the right hand side,
I Z
α · f (x)dl = −α · (∇ f (x) ∧ n) dA

or I Z 
α· f (x)dl + (∇ f (x) ∧ n) dA = 0

Since α is an arbitrary vector,

I Z
f (x)dl + (∇ f (x) ∧ n) dA = 0

4.4 Multipole expansion

We will demonstrate the multipole expansion for an electrostatic potential. If
ρ(x) is an arbitrary charge distribution then the potential due to it at a point
x, in gaussian units, is
ρ(x0 )
Z
φ(x) = dV (x0 )
|x − x0 |
If s = x − x0 then the Taylor expansion for s−1 is
x0i x0j ∂ 2 x0i x0j x0k ∂3
     
1 1 ∂ 1 1 1
= + x0i + + + ···
s r ∂xi s 2! ∂xi ∂xj s 3! ∂xi ∂xj ∂xk s
Therefore,
(2) (3)
Q(0) Qij ∂2 Qijk ∂3
     
(1) ∂ 1 1 1
φ(x) = + Qi + + +···
r ∂xi s 2! ∂xi ∂xj s 3! ∂xi ∂xj ∂xk s
where
Z
Q (0)
= ρ(x0 )dV (x0 )
Z
(1)
Qi = xi ρ(x0 )dV (x0 )
Z
(2)
Qij = xi xj ρ(x0 )dV (x0 )
Z
(3)
Qijk = xi xj xk ρ(x0 )dV (x0 )

112
are the multipoles of charge density. The zeroth order term Q(0) is the total
(1)
charge, the first order term Qi is the dipole moment, the second order term
(2) (3)
Qij is the quadrupole moment and Qijk is the octupole moment of the charge
distribution. If we write s2 = a (xa − x0a )2 then we observe that
P

xi − x0i
 
∂ 1
= −
∂xi s s3
∂2 (xi − x0i )(xj − x0j ) δij
 
1
= 3 − 3
∂xi ∂xj s s5 s
3 (xi − xi )(xj − xj )(xk − x0k )
0 0
 
∂ 1
= −15 +
∂xi ∂xj ∂xk s s7
3 
(xi − x0i )δjk + (xj − x0j )δik + (xk − x0k )δij

s5
Thus,
(2) (2)
Q(0)
(1)
(xi − x0i )Qi 3(xi − x0i )(xj − x0j )Qij − Qii
φ(x) = − + −
r s3 s5
(3) (3)
15(xi − x0i )(xj − x0j )(xk − x0k )Qijk + 3(xi − x0i )Qijj
+
  s7
1
O ,
s5
where we have used the symmetry of the octupole moment tensor to get
(3) (3) (3) (3)
(xi − x0i )Qijj + (xj − x0j )Qiji + (xk − x0k )Qiik = 3(xi − x0i )Qijj

The nth term of the expansion of φ goes as rn . We will not get an expression
for the field due to an arbitrary charge distribution. Since E = − ∇ φ,
−1
(xi − x0i )
 
(0) ∂(r ) (1) ∂
Ea = −Q + Qi −
∂xa ∂xa s3
3(xi − x0i )(xj − x0j ) − δij
 
(2) ∂
Qij +
∂xa s5
15(xi − x0i )(xj − x0j )(xk − x0k ) + 3(xi − x0i )δjk
 
(3) ∂
Qijk −
∂xa s7
···

Thus, the field due to a monopole goes as r−2 , that due to a dipole as r−3 ,
quadrupole as r−4 and so on.

4.5 Solid Harmonics

We consider the solution of Laplace’s equation using the method of separation
of variables. Let Φ(r, ϕ, θ) = R(r)S(ϕ, θ) be the solution of ∇2 Φ = 0. Writing
the Laplacian in spherical polar coordinates,
1 ∂2S
     
S d 2 dR R 1 ∂ ∂S
r + 2 sin θ + =0
r2 dr dr r sin θ ∂θ ∂θ sin2 θ ∂ϕ2

113
Multiplying both sides by r2 /Φ,

1 ∂2S
     
1 d 2 dR 1 1 ∂ ∂S
r =− sin θ +
R dr dr S sin θ ∂θ ∂θ sin2 θ ∂ϕ2
Since the left hand side is a function of r alone while the right hand side is a
function of ϕ and θ, each side is a constant. We write it as n(n + 1) in view of
the fact that the S is a spherical harmonic. Thus, the Laplace’s equation splits
into
 
1 d 2 dR
r = n(n + 1)
R dr dr
1 ∂2S
   
1 1 ∂ ∂S
sin θ + = −n(n + 1)
S sin θ ∂θ ∂θ sin2 θ ∂ϕ2
The second equation can be written as

1 ∂ 2 Sn
 
1 ∂ ∂Sn
sin θ + + n(n + 1)Sn = 0
sin θ ∂θ ∂θ sin2 θ ∂ϕ2
where we add a subscript n to S following the convention to indicate the eigen-
value. Since Sn is not a function of r, we can as well write it as

r2 ∇2 Sn + n(n + 1)Sn = 0

We can readily verify that R(r) = rn is a solution of

 
1 d dR
r2 = n(n + 1)
R dr dr

Thus, Φn (r, ϕ, θ) = rn Sn (ϕ, θ) is a solution of ∇2 Φ = 0. The functions Φn

are called solid harmonics. We will now show that if Φn solves the Laplace’s
equation then so does Φ−n−1 . To do so, let us find the conditions for rm Φn to
be a solution of the Laplace’s equation, where n is fixed. We observe that

∇2 (rm Φn ) = (m + n)(m + n + 1)rm+n−2 Sn − n(n + 1)rm+n−2 Sn

= m(m + 2n + 1)rm−2 Φn ,

where we used the fact that Sn is a spherical harmonic. Clearly, rm Φn is a

solution of Laplace’s equation only if m = −(2n + 1). Thus r−(2n+1) Φn =
r−(2n+1) rn Sn = r−n−1 Sn is a solution of Laplace’s equation.
The only spherically symmetric solutions which vanish at infinity are the
ones with n = −1. They are
A
Φ= ,
r
where A is a suitably chosen constant. If Φ is a solution of Laplace’s equation
then so are its derivatives. We consider only these solutions in the book.

114
4.6 Green function of Laplacian
4.6.1 Method 1
p
Let f (x1 , x2 , . . . , xn ) = g(r) where r = x21 + x22 + · · · + x2n . Therefore,

fx1 = g 0 (r)rx
x1
= g 0 (r) p 2 2
x1 + x2 + · · · + x2n
x21 1
fx1 x1 = g 00 (r) + g 0 (r) p 2 −
x21 + x22 + · · · + x2n x1 + x22 + · · · + x2n
x21
g 0 (r)
(x21 + x22 + · · · + x2n )3/2
x21 x22 + x23 + · · · + x2n
fx1 x1 = g 00 (r) 2 + g 0
(r)
x1 + x22 + · · · + x2n (x21 + x22 + · · · + x2n )3/2

Therefore,

g 0 (r)
fx1 x1 + fx2 x2 + · · · + fxn xn = g 00 (r) + (n − 1) p
x21 + x22 + · · · + x2n
or,
g 0 (r)
fx1 x1 + fx2 x2 + · · · + fxn xn = g 00 (r) + (n − 1)
r
If fx1 x1 + fx2 x2 + · · · + fxn xn = 0 then

g 0 (r) g 00 (r) 1
g 00 (r) = −(n − 1) ⇒ 0 = −(n − 1)
r g (r) r

which implies
ln g 0 (r) = −(n − 1) ln r + ln α,
where α is a constant of integration. Therefore g 0 (r)rn−1 = α. We solve this
ordinary differential equation for the following three cases
1. If n = 2, g(r) = α ln r + β and
2. If n 6= 2,
α
g(r) = + β,
rn−2
where α and β are constants of integration. For n ≥ 2, the function g is singular
at the origin. Therefore, it is also the Green function of Laplacian. If r = |x−x0 |
then (
0 α ln |x − x0 | + β if n = 2
G(x, x ) =
α|x − x0 |−(n−2) + β if n 6= 2

115
4.6.2 Method 2
Green function of a Laplacian is defined as
(
δ(x − x0 ) on V
∇2 G(x, x0 ) =
0 on S

where V is the volume enclosed by a surface S. When Neumann boundary

conditions are specified, the second condition is usually ∇ G(x, x0 ) · n = A−1 ,
where A is the area of a ‘sphere’ of unit radius centerd at x0 . To get the Green
function, we will integrate ∇2 G(x, x0 ) = δ(x − x0 ). In each case, the Green
function depends only on |x − x0 | because of the ‘spherical’ symmetry of the
problem.
1. In three dimensions,
Z Z
∇2 G(x, x0 )dV = δ(x − x0 )dV

or Z
∇· (∇ G(x, x0 ))dV = 1

Using divergence theorem,

Z
∇ G(x, x0 ) · ndS = 1

If the volume of integration were a sphere of radius r, since G is a function

of |x − x0 | alone, ∇ G(x, x0 ) = G0 (r)er . Further, since the unit normal n
to the sphere is in the radial direction, ∇ G(x, x0 ) · n = G0 (r). Hence,
Z
G0 (r)dS = 1 (4.6.1)

On a sphere, G0 (r) is constant, Therefore,

Z
G (r) dS = 1 ⇒ G0 4πr2 = 1
0

or
1
G(r) = − + c,
4πr
where c is a constant of integration. Since r = |x − x0 |,
1
G(|x − x0 |) = −
4π|x − x0 |

if we also choose c = 0.
2. In two dimensions, the treatment up to (4.6.1) is same as in the case of
three dimensions, except that the volume integral is really over a disk and
the surface integral is over a circle. Therefore,
Z
G0 (r) dS = 1 ⇒ G0 2πr = 1

116
 or
1
G(r) =
ln r + c,
2π
where c is a constant of integration. Since r = |x − x0 |,
1
G(|x − x0 |) = ln |x − x0 |
2π
if we also choose c = 0.
3. For n ≥ 3 dimensions, the equation (4.6.1) evaluates to

nπ n/2 rn−1
Z
G0 (r) dS = 1 ⇒ G0 = 1,
Γ(1 + n/2)

where we have used the result that the surface area of a n-sphere of radius
r is1
nπ n/2 rn−1
Sn (r) =
Γ(1 + n/2)
Therefore,
 n 1 1
G0 (r) = Γ 1 + n/2 n−1
2 nπ r
or  n  1 1
G0 (r) = −Γ 1 + n/2 n−2
+ c,
2 n(n − 2)π r
where c is a constant of integration. Since r = |x − x0 |,
 n 1 1
G(|x − x0 |) = −Γ 1 + (4.6.2)
2 n(n − 2)π n/2 |x − x0 |n−2

if we also choose c = 0. If n = 3, we get

1
G(|x − x0 |) = −
4π|x − x0 |

4. Interestingly, the (4.6.2) is valid even for n = 1, when we get

1
G(|x − x0 |) = |x − x0 |
2

4.6.3 Method 3
The definition of Green function is

∇2 G(x, x0 ) = δ (n) (x − x0 ) (4.6.3)

where n could be any dimension. Thus, in general,

n
X ∂2
∇2 ≡
∂x2k
k=1
1 Equation 18.77 in ‘Statistical Mechanics’ by K. Huang, 2nd edition.

117
Without loss of generality, choose x0 = 0 and take Fourier transform of (4.6.3)
to get
k 2 G̃(k) = 1
Thus, the Fourier transform of Green function of Laplacian of any dimensions,
in the momentum space is 1/k 2 . Finding an expression in position space is
just a matter of finding an inverse Fourier transform suitable for that space.
This technique works for all dimensions except n = 2. We know that in two
dimensions the Green function of a Laplacian is ln r. It diverges as r → ∞. The
theory of Fourier transforms needs the functions to be localized and decaying
to zero at infinity. Therefore, the usual techniques of integration do not work
in this case. Refer to the StackExchange conversation for more details2 .

4.7 Divergence theorem in a plane

Consider a vector field F(x, y) defined over a closed region S whose boundary
is denoted by ∂S. If ds denotes a small displacement along the boundary, the
tangent t and the normal n are defined as
dx dy
t ex +
= ey
ds ds
dy dx
n = ex − ey
ds ds
Refer to Mathworld for more details. The circulation of F = Lex + M ey is
defined as
I I   I
dx dy
F · tds = L +M ds = (Ldx + M dy)
ds ds
while the flux through ∂S is
I I   I
dy dx
F · nds = L −M ds = (Ldy − M dx)
ds ds
Green’s theorem in a plane is
I ZZ  
∂M ∂L
(Ldx + M dy) = − dxdy
∂x ∂y
Therefore,
I ZZ  
∂M ∂L
F · tds = − dxdy
∂x ∂y
I ZZ  
∂L ∂M
F · nds = + dxdy
∂x ∂y
However, the circulation is a line integral of the curl while the flux is a ‘surface’
integral of the field. Therefore, the above equations can as well be written as
I ZZ
F · tds = (∇ ∧ F )z dxdy
I ZZ
F · nds = (∇· F)dxdy
2 Dr. Robert E. Hunt of Cambridge University, in a private correspondence, explained this

point to me and directed me to the StackExchange conversation

118
4.8 Circular harmonics
Consider Laplace’s equation in plane polar coordinates for Φ(r, θ),
 
1 ∂ ∂Φ 1 ∂
r + 2 (Φ) = 0
r ∂r ∂r r ∂θ

Let Φ(r, θ) = R(r)S(θ) so that

1 d2 S
 
r d dR
r =−
R dr dr S dθ2

Since the left side depends on r alone while the right hand side depends on θ,
each side is equal to a constant, say m. Thus,

1 d2 S d2 S
− = m ⇒ + mS = 0
S dθ2 dθ2
Since S(θ) = S(θ+2π), we require m to be square of an integer. Let us therefore
write m = n2 . The R equation is
 
r d dR
r = m2
R dr dr

It is easy to check that if m2 6= 0 then R = rl is a solution while if m = 0,

R = ln r is a solution. Thus, the complete solution of Laplace’s equation is

Φ(r, θ) = (a0 + b0 ln r)(c0 + d0 θ) +

X
ak rk + bk r−k ck eikθ + dk e−ikθ ,

k6=0

where ak , bk , ck , dk are constants. Φ(r, θ) = Φ(r, θ + 2π) requires d0 = 0 so that,

X
ak rk + bk r−k ck eikθ + dk e−ikθ ,



Φ(r, θ) = (a0 + b0 ln r) +
k6=0

If the solution is required to be ‘circularly symmetric’, that is independent of

polar angle θ then we have to choose ak , bk , ck , dk = 0 for all k 6= 0, in which
case,
Φ(r, θ) = a0 + b0 ln r
Solutions of Laplace’s equation in plane polar coordinates are called circular
harmonics. In the book considers only those which are ‘circularly symmetric’.
Recall that if f (r) is a solution of Laplace’s equation then so are all deriva-
tives of f with respect to r.

4.9 A property of harmonic functions

Let φ1 (x) and φ2 (x) be two solution of Laplace’s equation. That it, ∇2 φ1 = 0
and ∇2 φ2 = 0. Consider

∇· (φ1 ∇ φ2 ) = ∇ φ1 · ∇ φ2 + φ1 ∇2 φ2 = ∇ φ1 · ∇ φ2 ,

119
because ∇2 φ2 = 0. Similarly,

∇· (φ2 ∇ φ1 ) = ∇ φ2 · ∇ φ1 + φ2 ∇2 φ1 = ∇ φ1 · ∇ φ2 ,

because ∇2 φ1 = 0. Therefore, ∇· (φ1 ∇ φ2 ) = ∇· (φ2 ∇ φ1 ). Taking a volume

integral of the equation,
Z Z
∇· (φ1 ∇ φ2 ) dV = ∇· (φ2 ∇ φ1 ) dV,

after using divergence theorem,

Z Z
φ1 ∇ φ2 · ndA = φ2 ∇ φ1 · ndA (4.9.1)

R
4.10 Proof of ni nj dΩ = (4π/3)δij
Consider the integral, Z
Iij = ni nj dΩ

where ni , nj are unit normals and dΩ denotes the solid angle. We will evaluate
this integral for the following sub-cases, which are all the ones that are possible.
• i = j = x. In this case, putting ni = nj = sin θ cos ϕ,
Z π Z 2π
Ixx = sin2 θ cos2 ϕ sin θdθdϕ
0 0

We observe that
Z 2π Z 2π  
1 + cos(2ϕ)
cos2 ϕdϕ = dϕ = π
0 0 2

and
Z π Z π
sin3 θdθ = (1 − cos2 θ) sin θdθ
0 0
Z −1
= − (1 − u2 )(−du)
1
Z 1
= (1 − u2 )du
−1
2 4
= 2− =
3 3
Therefore,
4π
Ixx =
3
• i = j = y. In this case, put ni = nj = sin θ sin ϕ, so that
Z π Z 2π
Iyy = sin2 θ sin2 ϕ sin θdθdϕ
0 0

120
 Since
2π 2π  
1 − cos(2ϕ)
Z Z
2
sin ϕdϕ = dϕ = π,
0 0 2
we get
4π
Iyy =
3
• i = j = z. In this case, put ni = nj = cos θ, so that
Z π Z 2π Z 1
2 4π
Izz = cos θ sin θdθdϕ = 2π u2 (−du) =
0 0 −1 3

• i = x, j = y. In this case, put ni = sin θ cos ϕ and nj = sin θ sin ϕ, so that

Z π Z 2π
Ixy = sin3 θdθ sin ϕ cos ϕdϕ
0 0

The ϕ integral is zero making Ixy = 0.

• i = y, j = z. In this case, put ni = sin θ sin ϕ and nj = cos θ, so that
Z π Z 2π
Iyz = sin2 θ cos θdθ sin ϕdϕ
0 0

The ϕ integral is zero making Iyz = 0.

• i = z, j = x. In this case, put ni = cos θ sin ϕ and nj = sin θ cos φ, so that

Z π Z 2π
Izx = sin2 θ cos θdθ cos ϕdϕ
0 0

The ϕ integral is zero making Izx = 0. The six cases above can be sum-
marized as Z
4π
ni nj dΩ = δij (4.10.1)
3
R
4.11 Proof of ni nj nk nl dΩ = (4π/15)(δij δkl +δik δjl +
δil δjk )
We will consider the following cases,
• All indices are the same. In this case, the integral is one of x4 , y 4 or z 4 ,
where x = sin θ cos ϕ, y = sin θ sin ϕ and z = cos θ. We first consider,
Z π Z 2π
Ixxxx = sin4 θ cos4 ϕ sin θdθdϕ
0 0
Z π Z 2π
5
= sin θdθ cos4 ϕdϕ
0 0

121
Now,
Z π Z π
sin5 θdθ = sin4 θ sin θdθ
0 0
Z π
= (1 − cos2 θ)5 sin θdθ
0
Z −1
= − (1 − 2u2 + u4 )du
1
16
=
15
and
Z 2π Z 2π  
4 3 cos 2ϕ cos 4ϕ
cos ϕdϕ = + + dϕ
0 0 8 2 8
3π
=
4
Therefore,
4π
Ixxxx =
5
Now consider,
Z π Z 2π
Iyyyy = sin4 θ sin4 ϕ sin θdθdϕ
0 0
Z π Z 2π
5
= sin θdθ sin4 ϕdϕ
0 0

We evaluate
Z 2π Z 2π  
3 cos 2ϕ cos 4ϕ
sin4 ϕdϕ = − + dϕ
0 0 8 2 8
3π
=
4
Therefore,
4π
Iyyyy =
5
Finally, consider
Z π Z 2π
Izzzz = cos4 θ sin θdθdϕ
0 0
Z π Z 2π
4
= cos θ sin θ dϕ
0 0
Z −1
= 2π u4 (−du)
1
4π
=
5
Thus, this point covers 3 cases.

122
• Three indices are same, one is different. This case includes integrands
of the form xyyy, xzzz, yzzz, yxxx, zyyy, zxxx. We evaluate each on of
them.
Z π Z 2π
Ixyyy = sin5 θdθ sin3 ϕ cos ϕdϕ
0 0
16
= ×0
15
= 0
Z π Z 2π
3 2
Ixzzz = cos θ sin θdθ cos ϕdϕ
Z0 π 0

= cos3 θ sin2 θdθ × 0

0
= 0
Z π Z 2π
Iyzzz = cos3 θ sin2 θdθ sin ϕdϕ
0 0
Z π
= cos3 θ sin2 θdθ × 0
0
= 0
Z π Z 2π
Iyxxx = sin5 θdθ sin ϕ cos3 ϕdϕ
0 0
16
= ×0
15
= 0
Z π Z 2π
3
Izxxx = sin θ cos θdθ sin3 ϕdϕ
0 0
Z 2π
= 0× sin3 ϕdϕ
0
= 0
Z π Z 2π
3
Izyyy = sin θ cos θdθ cos3 ϕdϕ
0 0
Z 2π
= 0× cos3 ϕdϕ
0
= 0

Note that
Ixyyy = Iyxyy = Iyyxy = Iyyyx = 0
and similarly for other cases. Thus, this point covers 6 × 4 = 24 cases.
• Two of them of one kind and two of the other. We therefore, consider the

123
 cases
Z π Z 2π
Ixxyy = sin5 θdθ sin2 ϕ cos2 ϕdϕ
0 0
2π
cos2 (2ϕ)
Z
16
= dϕ
15 0 4
16 2π 1 + cos(4ϕ)
Z
= dϕ
15 0 8
16 π
= ×
15 4
4π
=
15
Z π Z 2π
3 2
Ixxzz = sin θ cos θdθ cos2 ϕdϕ
0 0
Z π Z π  Z 2π
3 5 1 + cos(2ϕ)
= sin θdθ − sin θdθ dϕ
2
 0  0 0
4 16
= − π
3 15
4π
=
15
Z π Z 2π
Iyyzz = sin3 θ cos2 θdθ sin2 ϕdϕ
0 0
Z π Z π  Z 2π
1 − cos(2ϕ)
= sin3 θdθ − sin5 θdθ dϕ
0 0 0 2
 
4 16
= − π
3 15
4π
=
15

Note that
4π
Ixxyy = Ixyyx = Iyxyx = Ixyxy = Iyxxy = Iyyxx =
15
and similarly for other cases. Thus, this point covers 3 × 6 = 18 cases.
• Two of them of one kind and the remaining of two different kinds. We

124
 therefore, consider the cases
Z π Z 2π
4
Ixxyz = sin θ cos θdθ cos2 ϕ sin ϕdϕ
0 0
= 0×0
= 0
Z π Z 2π
Iyyxz = sin4 θ cos θdθ cos ϕ sin2 ϕdϕ
0 0
= 0×0
= 0
Z π Z 2π
2 3
Izzxy = sin θ cos θdθ cos ϕ sin ϕdϕ
Z0 π 0

= sin2 θ cos3 θdθ × 0

0
= 0

Note that

Ixxyz = Ixyzx = Ixyxz = Ixzyx = Ixzxy = Ixxzy = Iyzxx = Iyxxz =

Izxxy = Izxyx = Izyxx = Iyxzx

and similarly for the other cases. Thus, this point covers 3×12 = 36 cases.

The four points cover 3+24+18+36 = 81 cases, which are all the cases possible.
We can summarize all of them in the equation,
Z
4π
ni nj nk nl dΩ = (δij δkl + δik δjl + δil δjk ) (4.11.1)
15

4.12 Isotropic tensors

We will examine the nature of isotropic tensors of orders 0, 1, 2, 3 and 4. A
tensor is isotropic if it is invariant under a rotation of axes. The treatment in
this section follows that in H. Jeffrey’s ‘Cartesian Tensors’.

• A tensor of zero order is a scalar and it is invariant under a rotation of

coordinate axes.
• We will show that there are no isotropic tensors of order 1. A rotation of
coordinate axes by and angle θ about the z axis is represented by a the
orthogonal matrix,  
cos θ sin θ 0
− sin θ cos θ 0
0 0 1
For small angles of rotation, it can be approximated as
 
1 θ 0
−θ 1 0
0 0 1

125
 and decomposed as δij + cij , where the anti-symmetric matrix {cij } is
 
0 θ 0
−θ 0 0
0 0 0

We have thus shown that any small rotation can be represented by δij +
cij , where {cij } is an anti-symmetric matrix. A vector ui under a small
rotation is transformed to

u0j = (δij + cij )ui

It will remain unvariant under the rotation only if cij ui = 0. Writing in

full,

0.u1 + c12 u2 + c13 u3 = 0

−c12 u1 + 0.u2 + c23 u3 = 0
−c13 u1 − c23 u2 + 0.u3 = 0

Since  
 0 c12 c13 

−c12 0 c23  = 0

−c13 −c23 0
the only solution to cij ui = 0 is ui = 0. Thus, the only isotropic vector is
the null vector.
• Let uij be a second order tensor. Then u0kl = (δik − cik )uij (δjl − cjl ). Up
to first order in cij ,

u0kl = ukl − ukj cjl − uil cik

The tensor, uij will be isotropic only if ukj cjl + uil cik = 0. If k 6= l, choose
k = 1 and l = 2 so that u1j cj2 + ui2 ci1 = 0 or

u11 c12 + u12 c22 + u12 c32 + u12 c11 + u22 c21 + u32 c31 = 0

Using symmetry properties of {cij },

u11 c12 − u12 c23 − u22 c12 − u32 c13 = 0

or
(u11 − u22 )c12 − u12 c23 − u32 c13 = 0
This can be true, in general, only if u11 = u22 and u12 = u32 = 0.
Similarly, by choosing k = 1 and l = 3, we get u11 = u33 and u12 = u23 =
0. Further, choosing k = 2 and l = 3 gives u22 = u33 and u21 = u13 = 0.
If we choose k = l = 1, u1j cj1 + ui1 ci1 = 0 or

u11 c11 + u12 c21 + u13 c31 + u11 c11 + u21 c21 + u31 c31 = 0

Using symmetry properties of {cij },

(u12 + u21 )c21 + (u13 + u31 )c31 = 0

126
 which can be true, in general, only if u12 = −u21 and u13 = −u31 . But
we have already shown that u12 = 0 and u13 = 0. Thus, u21 and u31 also
vanish. Choosing k = l = 3 we can show that u23 + u32 = 0, which since
u23 = 0 implies vanishing of both. We have thus shown that the only
isotropic tensor is a multiple of the Kronecker delta δij .
• Let uijk be a third order tensor. Then

u0lmn = (δil − cil )(δjm − cjm )(δkn − ckn )uijk

Up to first order in cij ,

u0lmn = ulmn − δil δjm ckn uijk − δil δkn cjm uijk − δjm δkn cil uijk ,

or
u0lmn = ulmn − ckn ulmk − cjm uljn − cil uimn
The tensor uijk will be isotropic only if u0lmn = ulmn or, if

ckn ulmk + cjm uljn + cil uimn = 0

If l = m = 1, the above equation becomes ckn u11k + cj1 u1jn + ci1 ui1n = 0.
Summing it over all indices, except n, we get

c1n u111 + c2n u112 + c3n u113 + c11 u11n + c21 u12n +
c31 u13n + c11 u11n + c21 u21n + c31 u31n = 0

Using symmetry properties of {cij },

c1n u111 + c2n u112 + c3n u113 + c21 (u12n + u21n ) + c31 (u13n + u31n ) = 0

In the particular case of n = 2, we get

c12 (u111 − u122 − u212 ) + c32 u113 + c31 (u132 + u312 ) = 0

This can be true, in general, only of

u122 + u212 = u111

u132 + u312 = 0
u113 = 0

From the last of the above three equations, uijk = 0 if any two of the
three indices are same. This fact, coupled with the first of the three
equations tells that uijk = 0 if all indices are equal. Finally, the second
of the above equations tells that uijk = −ujik . Thus, the tensor entry for
(i, j, k) changes sign under an odd permutation of indices. Thus, the only
isotropic tensor of third order is a scalar multiple of Levi-Civita tensor
ijk .
• Let uijkl be a fourth order tensor and let

u0pqrs = (δip − cip )(δjq − cjq )(δkr − ckr )(δls − cls )uijkl

127
Then, up to first order in cij ,

u0pqrs = δip δjq δkr δls uijkl − cip δjq δkr δls uijkl
−δip cjq δkr δls uijkl − δip δjq ckr δls uijkl − δip δjq δkr cls uijkl
= upqrs − cip uiqrs − cjq upjrs − ckr upqks − cls upqrl

The tensor uijkl will be isotropic only if

cip uiqrs + cjq upjrs + ckr upqks + cls upqrl = 0 (4.12.1)

There are three possible values for the four indices i, j, k, l. They give rise
to four possibilities:
1. Two indices are equal and the other two are unequal. Choose p =
q = 1, r = 2 and s = 3 so that (4.12.1) becomes

ci1 ui123 + cj1 u1j23 + ck2 u11k3 + cl3 u112l = 0

which is same as

c11 u1123 + c21 u2123 + c31 u3123 + c11 u1123 + =0

c21 u1223 + c31 u1323 + c12 u1113 + c22 u1123 +
c32 u1133 + c13 u1121 + c23 u1122 + c33 u1123

Using symmetry properties of {cij },

c12 (u1113 − u2123 − u1223 ) + c13 (u1121 − u3123 − u1323 )+ = 0

c23 (u1122 − u1133 )

This equation will be valid for an arbitrary cij only if

u1113 − u2123 − u1223 = 0 (4.12.2)

u1121 − u3123 − u1323 = 0 (4.12.3)
u1122 − u1133 = 0 (4.12.4)

The last of the above equations gives u1122 = u1133 . Choosing p =

r = 1, q = 2 and s = 3 will give u1212 = u1313 . Other combinations
will similar identities. In general,

u1122 = u1133 = u2211 = u2233 = u3311 = u3322 (4.12.5)

u1212 = u1313 = u2121 = u2323 = u3131 = u3232 (4.12.6)

u1221 = u1331 = u2112 = u2332 = u3113 = u3223 (4.12.7)
Thus, all terms of uijkl where two indices are of one kind and the
remaining two are of the other kind are equal.
2. Three indices are equal and the remaining one is different from them.
Choose p = q = r = 1 and s = 2 so that (4.12.1) becomes

ci1 ui112 + cj1 u1j12 + ck1 u11k2 + cl2 u111l = 0

128
 which is same as

c11 u1112 + c21 u2112 + c31 u3112 + c11 u1112 + =0

c21 u1212 + c31 u1312 + c11 u1112 + c21 u1122 +
c31 u1132 + c12 u1111 + c22 u1112 + c32 u1113

Using symmetry properties of {cij },

c12 (u1111 − u2112 − u1212 − u1122 )− = 0

c13 (u3112 + u1312 + u1132 ) − c23 u1113

This equation can be true for arbitrary cij only if

u1111 = u2112 + u1212 + u1122 (4.12.8)

0 = u3112 + u1312 + u1132 (4.12.9)
0 = u1113 (4.12.10)

The last of the above equations tells that all terms of uijkl where
three indices are same and one differs from the rest are zero. Using
this fact in (4.12.2) and (4.12.3), we get

u2123 + u1223 = 0 (4.12.11)

u3123 + u1323 = 0 (4.12.12)

(4.12.11) under the transformation 1 7→ 3, 2 7→ 1, 3 7→ 2 becomes

u1312 + u3112 = 0,

which put with (4.12.9) gives u3112 = 0. By symmetry, all compo-

nents of uijkl whose two indices are same and the rest two are different
are zero. Further, equation (4.12.8) tells that all components of uijkl ,
all whose indices are equal, are written as sums of components two
of whose indices are of one value and the rest are of other value.
To summarize:
– Components of uijkl where three indices are same and one differs
from the rest are zero.
– Components of uijkl whose two indices are same and the rest two are
different are zero.
– Components of uijkl , all whose indices are equal, are written as sums
of components, two of whose indices are of one value and the rest are
of other value.
– Components of uijkl , two of whose values are equal and rest are of
other value satisfy equations (4.12.5), (4.12.6) and (4.12.7).
Let
u1122 = u1133 = u2211 = u2233 = u3311 = u3322 = γ
u1212 = u1313 = u2121 = u2323 = u3131 = u3232 = a
u1221 = u1331 = u2112 = u2332 = u3113 = u3223 = b,

129
where γ, a, b are scalars. For sake of convenience, we can introduce two
new scalars
a+b
α =
2
a−b
β =
2
so that
u1122 = u1133 = u2211 = u2233 = u3311 = u3322 = γ (4.12.13)
u1212 = u1313 = u2121 = u2323 = u3131 = u3232 = α + β (4.12.14)
u1221 = u1331 = u2112 = u2332 = u3113 = u3223 = α − β (4.12.15)
From (4.12.8),
u1111 = (α − β) + (α + β) + γ = γ + 2α
By symmetry, we conclude that
u1111 = u2222 = u3333 = γ + 2α (4.12.16)

From (4.12.14) and (4.12.15), we observe that u1212 = γ + β and u1221 =

γ − β so that
u1212 + u1221 = 2α
u1212 − u1221 = 2β
Thus, there are two tensors, uijij + uijji contributing 2α and uijij − uijji
contributing 2β. Equivalently, we can say that the tensor wijkl is such
that wijij and wijji each contribute α and the other choices of indices
contribute zero, while the tensor vijkl is such that vijij contributes β, vijji
contributes −β and the other choices of indices contribute zero.
Equations (4.12.13), (4.12.14) and (4.12.15) suggest that there are three
independent isotropic tensors described as
1. uijkl = 1 if i = j and k = l and zero other otherwise. It can be
represented as γδij δkl .
2. uijkl = 1 if i = k and j = l or i = l and j = k and i 6= j. uijkl = 0 for
all other choices of indices. Further, if all indices are same uijkl = 2.
It can be represented as α(δik δjl + δil δjk).
3. uijkl = 1 if i = k and j = l, uijkl = −1 if i = l and j = k and
uijkl = 0 for all other choices of indices. It can be represented as
β(δik δjl − δil δjk).
Therefore, a general isotropic tensor of 4th order is expressed as
uijkl = γδij δkl + α(δik δjl + δil δjk ) + β(δik δjl − δil δjk ) (4.12.17)
An equivalent form is
uijkl = µ00 δij δkl + µδik δjl + µ0 δil δjk
or,
uijkl = µδik δjl + µ0 δil δjk + µ00 δij δkl (4.12.18)

130